CROP RECOMMENDATION SYSTEM USING

MACHINE LEARNING

A PROJECT REPORT

Submitted by

LIBANYA GIFTY T (950720112052)

in partial fulfilment for the award of the minor degree of

BACHELOR OF TECHNOLOGY

in

INFORMATION TECHNOLOGY

FRANCIS XAVIER ENGINEERING COLLEGE

(Autonomous)

TIRUNELVELI –627 003

APRIL 2025
FRANCIS XAVIER ENGINERING COLLEGE (Autonomous)
TIRUNELVELI 627 003

BONAFIDE CERTIFICATE

Certified that his project report “CROP RECOMMENDER SYSTEM USING

MACHINE LEARNING” is the Bonafide work of “LIBANYA GIFTY T
(950720112052)” who carried out the project work under my supervision. Certified
further that to the best of my knowledge the work reported herein does not form part of
any other thesis or dissertation on the basis of which a degree or award was conferred
on an earlier occasion on this or any other candidate.

MRS. M. VINOTHINI, M.E., Dr. M. CAROLINE VIOLA STELLA MARY, M.E., Ph.D.,

SUPERVISOR HEAD OF THE DEPARTMENT

Associate Professor Professor and Head

Department of Information Technology Department of Information Technology

Francis Xavier Engineering College Francis Xavier Engineering College

Tirunelveli. Tirunelveli.

Submitted for the B. Tech MINOR Degree Project Viva Voce held on ………………………………

INTERNAL EXAMINER EXTERNAL EXAMINER

ii
 ACKNOWLEDGEMENT

It First and foremost, we praise and thank “THE GOD

ALMIGHTY” the Lord of all creation, who by His abundant grace has
sustained us and helped us to complete the project successfully. We really
find unique and immense pleasure in thanking our respected Chairman of
our college Dr. S. Cletus Babu M.A., Ph.D. A deep bouquet of thanks to
respected Principal Dr. V. Velmurugan M.E., Ph.D., for having
provided the faculties required for pursuing our project.

We sincerely thank Dr. M. Caroline Viola Stella Mary M.E., Ph.D.,

Professor and Head Department of Information Technology, Francis
Xavier and Engineering college and project coordinator Ms. R. Shirly
Myrtle M.E., Assistant Professor, Department of Information
Technology, who inspired us and gave us time to make this project to work
a grant success.
We also thank our guide Assistant Professor MRS. M.
VINOTHINI, M.E., Department of Information Technology, for her
valuable guidance throughout and it is our great privilege to express our
gratitude to him. We extend our heartfelt thanks and profound gratitude to
all the faculty members of Information Technology Department for their
kind help during our project work.
Finally, we express my sincere thanks to our parent, who have
constantly encouraged us and throughout our course.

iii
 Abstract

Agriculture plays a crucial role in food security and economic growth, making

efficient crop selection essential for maximizing yield and promoting

sustainability. The Crop Recommendation System leverages machine learning to

assist farmers in identifying the most suitable crops based on soil characteristics

and climatic conditions. The system takes Nitrogen, Phosphorus, Potassium,

Temperature, Humidity, pH, and Rainfall as input parameters, standardizes them

using a scaling technique, and predicts the optimal crop using a pre-trained model.

Built using Flask, this web-based application provides an easy-to-use interface

for users to input their data and receive recommendations. By utilizing historical

agricultural data, it promotes sustainable farming by ensuring optimal resource

utilization and reducing reliance on trial-and-error methods. The system has the

potential to minimize input costs and maximize crop yield, directly contributing

to increased farmer income It also fosters environmental conservation by

recommending crops that suit existing soil and climatic conditions. By using

historical data, predictive models, and a user-friendly web interface, the system

empowers farmers to make data-driven, informed decisions that enhance

productivity, profitability, and sustainability. In the ongoing journey to digitalize

agriculture, such AI-based solutions will play a crucial role in creating a more

resilient and efficient farming ecosystem. Additionally, the development of a

mobile application will further improve accessibility and effectiveness, benefiting

agricultural communities worldwide.

iv
 TABLE OF CONTENTS

CHAPTER PAGE
TITLE
NO NO
ACKNOWLEDGEMENT iii
ABSTRACT iv
LIST OF FIGURES viii
LIST OF ABBREVIATIONS ix

1. INTRODUCTION
1.1 General Introduction 10
1.2 Project Objective 11
1.3 Project Goal 12
1.4 Project Scope 13
1.5 Project Purpose 13

2. LITERATURE REVIEW
2.1 Introduction 15
2.2 Literature Review 16

2.3 Summary 20

3. EXISTING SYSTEM
3.1 Introduction to Existing Crop Recommendation Systems 22
3.1.1 Evolution of Crop Recommendation Technologies 22
3.1.2 Importance of Crop Recommendation Systems 23
3.1.3 Features of Existing Systems 23
3.1.4 Accessibility and Platform Integration 23
3.1.5 Integration and Usability 24
3.2 Analysis of Existing Methodology 24
3.2.1 Data Collection Methods 24
3.2.2 Data Processing and Analysis 25
3.2.3 User Interaction and Experience 25
3.3 Case Studies or Examples 25

v
 3.3.1 Random Forest-Based Crop Recommender 25
3.3.2 KNN-Based Localized Crop Predictor 26
3.4. Challenges Faced by Existing Systems 26
3.4.1 Data quality and completeness 26
3.4.2 Computational Constraints 26
3.4.3 Usability and Accessibility 27
3.4.4 Generalization and Model Adaptability 27

4. PROPOSED SYSTEM
4.1 Introduction 28
4.2 Motivation 29
4.3 Objectives and Goals 30
4.4 System Architecture 32

4.4.1 Architecture Overview 33

4.4.2 Crop Prediction Model Module 34
4.4.3 Workflow and Interaction Between Components 35
4.4.4 Visual Representation 37
4.5 Implementation Details 39
4.5.1 Technology Stack 39
4.5.2 Key Features 40
4.5.3 Development Methodology 41
4.5.4 Challenges Faced During Development 43
4.6 Testing and Evaluation 44
4.6.1 Testing Methodologies 44
4.6.2 Evaluation Criteria 45
4.6.3 Results and Findings 46
4.6.4 Discussion and Recommendations 47

5. RESULT AND DISCUSSION

5.1 Introduction 48
5.2 Real-Time Crop Recommendation Process 48
5.3 Accuracy and Performance Analysis 51

vi
 5.4 Performance in Different Environments 52

6. CONCLUSION AND FUTURE WORK

6.1 Conclusion 53
6.2 Future Work 53

APPENDIX 1 - PROGRAM CODE 55

7. APPENDIX 2 - SCREENSHOTS 59
REFERENCES 61

vii
 LIST OF FIGURES
FIGURE PAGE NO
NO TITLE

1 System Architecture 27

2 Data Flow Diagram 31

3 Flow Chart 32

4 Enter the values of quantity of soil 43

5 Recommended crops 44

6 Power BI Average Dashboard 45

viii
 LIST OF ABBREVIATIONS

ACRONYM ABBREVIATION

API Application Programming Interface

ML Machine Learning

PKN Phosphorus Potassium Nitrogen

RF Random Forest

SVM Support Vector Machine

UI User Interface

ix
 CHAPTER 1
INTRODUCTION

1.1 INTRODUCTION

In agriculture and data science, intelligent decision-making systems are being

widely used to improve crop productivity and sustainability. Crop recommendation
is one such significant application that helps recommend the most appropriate crop
to grow based on different environmental and soil factors. Through machine learning
and data analytics, farmers and agricultural specialists can make better decisions that
enhance yield, minimize wastage of resources, and ensure sustainable farming.

This project, "Crop Recommendation System using Machine Learning," aims

to create an intelligent system that evaluates critical agricultural characteristics like
soil type, temperature, humidity, rainfall, and nutrient levels (NPK - Nitrogen,
Phosphorus, Potassium) in order to suggest the most suitable crop for farming. With
the help of machine learning algorithms and data preprocessing, the system can
recognize intricate patterns and relationships between the data, resulting in more
precise and individualized crop recommendations.

The report gives a detailed description of the development process, ranging

from data gathering, feature engineering, model choice, and evaluation. It also points
out the difficulties faced and how they were overcome during the implementation.
The system is developed using Python and machine learning libraries like Scikit-
learn, Pandas, and NumPy, while insights are represented graphically using Power
BI for better interpretability and decision-making support.

10
 This report documents a detailed overview of the project, including the
approaches used, technologies and tools employed, and issues faced during
development. It also reflects on the use of machine learning for interpreting
agricultural data to deliver useful recommendations, highlighting the role of AI-
based decision-making in contemporary farming practices.

1.2 PROJECT OBJECTIVE

The goal of this project is to develop and design an intelligent, automated crop
recommender system based on machine learning methods. The system should help
farmers and stakeholders in agriculture by predicting the best crop to be grown
accurately from different environmental and soil parameters like Nitrogen (N),
Phosphorus (P), Potassium (K) levels, temperature, humidity, pH of the soil, and
rainfall. This method greatly minimizes guesswork and improves agricultural
decision-making, leading to greater productivity and optimization of resources.

This project aims to introduce data-driven, real-time intelligence to the

agricultural process, making it useful for use in precision farming, smart agriculture,
and sustainable crop management. Scalability and flexibility to various regions and
types of soil are ensured through the ability to retrain or fine-tune the model with
local data. Built using Python and important libraries like Scikit-learn, Pandas, and
NumPy, the model is able to process quickly and make efficient predictions.

In addition, by incorporating Power BI for interactive dashboards and visual

analytics, the system improves user experience through clear, data-driven insights
presented in an easily accessible manner. This encourages informed decision-making
and supports widespread adoption among farmers, agronomists, and policymakers
seeking to modernize and optimize agricultural practices.

11
1.3 PROJECT GOAL

The main objective of this project is to implement a precise, effective, and

scalable crop recommendation system based on machine learning to aid in smart
agricultural decision-making. Based on the critical soil and environmental factors
like Nitrogen (N), Phosphorus (P), Potassium (K), temperature, humidity, pH, and
rainfall, the system should suggest the ideal crop to cultivate under certain
conditions. This minimizes dependence on conventional practices and makes
farmers more data-driven in informing decisions for maximization of crop yield and
usage of resources.

One of the primary goals is to provide high prediction accuracy and efficient
inference, supporting real-time recommendations even in low-resource settings. The
machine learning model will be trained on varied agricultural datasets, enabling it to
learn intricate relationships between weather and soil attributes and crop suitability.
This flexibility guarantees consistent performance across various geographic
locations and seasonal changes.

In addition to improving the system's efficiency, data preprocessing methods

will be used to cleanse and normalize inputs prior to entering them into the model.
Techniques like Random Forest or Support Vector Machines (SVM) will be
researched and optimized for the best accuracy. The backend will be done using
Python, utilizing libraries such as Scikit-learn, Pandas, and NumPy for effective
model training and testing.

By providing a strong, smart, and easy-to-use crop suggestion system, this

project will encourage precision farming, enhance food security, and enable
sustainable agriculture through the capabilities of AI and machine learning.

12
1.4 PROJECT SCOPE

The purpose of this project is to automate crop recommendation based on

machine learning to give effective and reliable suggestions according to the
environment and soil conditions. The system is optimized to process structured input
data like Nitrogen (N), Phosphorus (P), Potassium (K), temperature, humidity, pH,
and rainfall—thus fitting a broad array of agricultural use cases, ranging from small-
scale farms to commercial agricultural operations.

The system is based on supervised machine learning algorithms trained on

extensive agricultural datasets to predict the most appropriate crop to be grown under
specified conditions. It integrates data preprocessing techniques for dealing with
inconsistencies, normalizing values, and improving input data quality. Pandas,
NumPy, and Scikit-learn libraries are employed to execute models such as Decision
Trees and Random Forest that have high accuracy as well as interpretability
capabilities in classification problems.

Its modular and scalable design guarantees the system can be scaled or
configured for varying regions, crop types, or other parameters. Altogether, this
project facilitates smart and sustainable agriculture through decreased guesswork,
optimized resource allocation, and enhanced crop yield utilizing sophisticated
machine learning and data science.

1.5 PROJECT PURPOSE

The main aim of the Crop Recommendation System via Machine Learning is
to create a smart, effective, and self-sustained system for recommending the most
appropriate crops under certain environmental and soil conditions. Crop choice is a

13
key decision in agriculture that has direct implications on yield, sustainability, and
resource utilization.

The system accepts input parameters like Nitrogen (N), Phosphorus (P),
Potassium (K), temperature, humidity, pH level, and rainfall. The inputs are given
to a machine learning model trained on the inputs, which inspects patterns and
suggests the best crop to cultivate. The outputs are given in a formatted way,
providing both the suggested crop and the data to support it so that the users can
know why the suggestion is made.

The project illustrates the use of Python for model building, Scikit-learn for
applying machine learning algorithms, and data visualization libraries to improve
interpretability. By making the crop recommendation system automated, the system
enables precision farming, enabling farmers to take well-informed decisions,
minimize the risk of crop failure, and increase productivity in general.

14
 CHAPTER 2

LITERATURE SURVEY

2.1 INTRODUCTION

Crop recommendation systems have gained significant attention in the domain

of smart agriculture, aiming to enhance productivity and sustainability through data-
driven insights. Traditional methods of crop selection based on experience and local
knowledge often lead to sub-optimal yields due to unaccounted variables like soil
health, climatic conditions, and rainfall patterns.

Earlier approaches to crop selection relied on human expertise, empirical data,

and local environmental knowledge, often leading to inconsistent yields due to
overlooked variables.

With the rise of data analytics and AI, machine learning models now allow
for the integration of multi-dimensional data such as temperature, rainfall, humidity,
and soil pH to make accurate and scalable predictions.

With advancements in machine learning, especially supervised learning

algorithms, the development of predictive models has become feasible. These
models analyze historical agricultural data and environmental parameters to suggest
the most suitable crops for cultivation. Various techniques, such as Decision Trees,
Random Forests, Support Vector Machines (SVM), and K-Nearest Neighbours
(KNN), have been employed for this purpose.

This literature survey reviews existing research in crop recommendation

systems and highlights the contributions of machine learning in transforming
conventional farming into intelligent, predictive agriculture.
15
2.2 LITERATURE REVIEW

With the growing complexity of climate patterns and soil degradation,

traditional farming methods are increasingly insufficient in optimizing crop yield.
As a result, machine learning (ML) has emerged as a reliable approach to enhancing
decision-making in agriculture. Specifically, ML-based crop recommendation
systems leverage historical and real-time data—such as soil nutrients, temperature,
humidity, and rainfall—to suggest the most suitable crops for cultivation. This
literature review explores various research studies that focus on data-driven
solutions in agriculture, emphasizing crop prediction, soil analysis, and smart
farming technologies.

Patil, R., & Deshmukh, A. – Machine Learning-Based Crop Recommendation

System (2022)

Kumar and Mehta (2023) discuss the transformative impact of artificial

intelligence on agriculture, particularly through the adoption of precision farming
techniques. Their research focused on leveraging machine learning algorithms such
as K-Nearest Neighbours (KNN) and Support Vector Machines (SVM) to develop a
system capable of recommending the most suitable crops based on varying
environmental and soil parameters. By analysing inputs like temperature, rainfall,
humidity, and essential soil nutrients, the system classified optimal crops for specific
conditions. The study demonstrated that AI-powered crop prediction models
significantly improved agricultural yield—by as much as 22%—especially in rural
farming communities where traditional farming practices often rely on outdated
knowledge and lack scientific guidance. The results validated the effectiveness of
data-driven decision-making in enhancing farming efficiency and minimizing the
risks associated with crop failure. Furthermore, the researchers emphasized the

16
scalability and adaptability of their system, suggesting that with the integration of
real-time sensor data and mobile platforms, such AI-based tools could be deployed
widely, even in remote or resource-limited areas. This research underscores the
potential of machine learning not only to enhance crop productivity but also to
empower farmers with timely, location-specific recommendations for sustainable
agriculture.

Kumar, S., & Mehta, R. – Application of AI in Precision Agriculture (2023)

The authors highlight the role of artificial intelligence in transforming

traditional agriculture into precision farming. Their system utilized K-Nearest
Neighbours (KNN) and Support Vector Machines (SVM) to classify optimal crops
based on weather and soil features. Results showed that AI-powered predictions
helped improve crop yield by 22%, making it a powerful tool for rural farmers.

Rao, H., & Sen, D. – Multi-Factor Crop Recommendation Using Ensemble

Learning (2022)

Rao and Sen focused on enhancing crop prediction by incorporating multi-

factor inputs, including soil nutrient content, previous crop patterns, and market
demand indicators. They employed ensemble learning techniques like Gradient
Boosting and Random Forest to generate more accurate crop suggestions. The
study concluded that multi-factor models outperform single-input systems by
capturing complex interactions among variables, leading to a 17% increase in yield
accuracy. This approach also introduced economic factors into crop
recommendations, suggesting crops based not only on suitability but also
profitability, a direction that could help farmers align agricultural planning with
market trends.

17
Patel, V., & Bhavsar, P. – Cloud-Based Crop Advisory Using Machine Learning
(2021)

This research introduced a cloud-integrated platform for real-time crop

advisory using machine learning algorithms such as Logistic Regression and
Decision Tree Classifier. Farmers could input soil parameters through a mobile
interface, and the system would analyze the data via a backend ML engine hosted on
cloud servers. This model showed scalability across regions, with minimal
infrastructure requirements at the user end. The research highlighted how cloud
computing can bridge the digital divide, offering smart farming tools even in
bandwidth-limited rural areas.

Singh, R., & Das, A. – Role of Feature Selection in Crop Prediction Accuracy
(2022)

Singh and Das analysed the impact of different feature selection techniques
on the performance of crop recommendation systems. Techniques such as
Recursive Feature Elimination (RFE) and Principal Component Analysis (PCA)
were applied to refine datasets. Their study found that proper feature engineering
significantly improved model performance, especially in algorithms like SVM and
Boost. By reducing irrelevant features, they managed to enhance both speed and
accuracy, showing that intelligent feature selection is crucial in real-time
deployment of recommendation system

Fernandez, M., & Rao, S. – Integration of GIS and Machine Learning in Crop
Suitability for the Mapping and analysing system in the analysis (2022)

In this study, Geographic Information Systems (GIS) were combined with

machine learning models to produce visual crop suitability maps. Data layers such

18
as land slope, elevation, and soil type were processed using Decision Tree classifiers
to generate crop-specific zoning maps. This visual representation provided intuitive
and localized decision support for farmers and policymakers. The study proposed its
integration into national e-governance platforms to enhance precision agriculture at
the policy level.

Zhang, L., & Prakash, R. – Real-Time Crop Forecasting Using LSTM Models
(2023)
This study introduced the use of Long Short-Term Memory (LSTM) networks
for crop yield forecasting and recommendation based on weather time-series data.
Unlike traditional models that rely on static input, LSTM networks capture temporal
dependencies, making them effective in forecasting future conditions. The system
was tested across three agricultural zones and showed consistent results in adapting
to climate variability. This research demonstrated the potential of deep learning
models in long-term agricultural planning.

Banu, H., & Krishnan, A. – Mobile-Based Intelligent Crop Recommendation

System (2023)

The authors developed a mobile application that uses built-in GPS and
climate APIs to fetch location-based weather conditions in real time. Coupled with
a lightweight ML model trained on regional datasets, the app provides crop
recommendations instantly based on both real-time and historical data. Field trials
indicated that the mobile-based system was effective in reaching underserved
farmers, especially those without access to desktop computing or consistent internet.

19
Ali, M., & Zhou, X. – Integrating Remote Sensing and ML for Agricultural
Planning (2023)

The authors developed a crop recommendation system that integrates satellite-

based remote sensing data with machine learning algorithms. By analysing
vegetation indices and soil moisture levels from satellite imagery, the system
recommends crops best suited for large-scale farming. The research emphasizes the
role of big data in future agricultural technologies.

Joshi, A., & Nair, V. – Smart Farming Using IoT and ML (2022)
Combining IoT-based sensors with machine learning, this research developed
a dynamic crop recommendation system that adapts to real-time changes in soil
moisture and weather. The system utilizes cloud computing to store and process large
volumes of sensor data and updates crop recommendations accordingly. The project
demonstrated how real-time feedback could significantly reduce water and fertilizer
waste.

2.3 SUMMARY

Crop recommendation systems have become increasingly important in

addressing the challenges of modern agriculture, including climate variability,
declining soil fertility, and inefficient resource usage. Researchers have explored
various techniques to improve the accuracy and adaptability of these systems by
incorporating diverse environmental and soil parameters. Traditional agricultural
practices often relied on heuristic knowledge and broad generalizations, which could
not account for local soil differences, weather fluctuations, or nutrient deficiencies.
Early decision-support systems offered basic recommendations based on soil type or

20
crop calendars but lacked the adaptability and precision required for optimal yield
outcomes.

Random Forest and Boost emerged as powerful tools capable of handling

multi-dimensional agricultural datasets. These models can analyze features such as
nitrogen (N), phosphorus (P), potassium (K), soil pH, temperature, humidity, and
rainfall to generate precise crop suggestions tailored to specific geographies.

In recent years, the integration of advanced technologies such as IoT sensors,

real-time data feeds, and deep learning has further enhanced the accuracy and
responsiveness of these systems. Convolutional Neural Networks (CNNs) and Long
Short-Term Memory (LSTM) models are now being used to analyze temporal and
image-based agricultural data, predicting seasonal crop trends and detecting early
signs of crop stress. Our project leverages supervised machine learning techniques
and data visualization tools to provide an intelligent and accessible crop
recommendation platform. Built using Python, Scikit-learn, and integrated with
Power BI dashboards, the system supports real-time analysis, interpretable results,
and scalable deployment. These technological advancements are transforming crop
planning into a data-driven, sustainable practice that supports food security,
environmental protection, and economic stability in the agricultural sector.

21
 CHAPTER 3

EXISTING SYSTEM

3.1. INTRODUCTION

Crop recommendation systems are designed to guide farmers in selecting the

most suitable crop for cultivation based on soil and environmental parameters. These
systems utilize techniques ranging from rule-based approaches to advanced machine
learning algorithms to provide accurate and data-driven crop suggestions. Early
systems primarily depended on generalized agricultural advisories or traditional
knowledge, which lacked precision and adaptability to local soil and weather
conditions. Recent developments in artificial intelligence have enabled the design of
more responsive and location-specific systems, significantly improving productivity
and decision-making in agriculture.

3.1.1 EVOLUTION OF CROP RECOMMENDATION TECHNOLOGIES

The evolution of crop recommendation has followed the technological

transformation in data science and agriculture. Initial systems relied on static
datasets and expert-defined rules to make crop suggestions. These methods were
limited in scale, flexibility, and responsiveness to real-time changes. With the
emergence of machine learning, data-driven systems began analysing
multidimensional factors—such as nitrogen (N), phosphorus (P), potassium (K), pH
level, temperature, humidity, and rainfall—to predict the best-suited crop for specific
conditions. Today, advanced algorithms like Random Forest, Decision Trees, and K-
Nearest Neighbours (KNN) enable precise crop classification based on historical
data and real-time inputs, with models being integrated into mobile apps and web
platforms for accessibility.
22
3.1.2 IMPORTANCE OF CROP RECOMMENDATION SYSTEMS

Modern agriculture demands intelligent systems to enhance crop productivity

while ensuring environmental sustainability. Crop recommendation systems reduce
the risk of crop failure, optimize the use of inputs such as fertilizers and water, and
support better land utilization. They also assist governments and agricultural
organizations in planning food supply chains and addressing regional farming
challenges. These systems empower farmers with timely, location-specific advice,
minimizing dependency on guesswork or outdated agricultural calendars.

3.1.3 FEATURES OF EXISTING SYSTEMS

Many existing systems are built using machine learning algorithms trained on
agricultural datasets. These systems typically allow users to input values such as
soil nutrient content and weather parameters, and receive real-time crop
suggestions. Some systems incorporate dynamic weather data through APIs or IoT
sensor integrations. Key features include:
• Predictive crop recommendations based on soil test results.
• Integration with real-time weather data for adaptive suggestions.
• Interactive dashboards for visualizing crop suitability and trends.
• Support for multiple crops and soil types across different regions.

3.1.4 ACCESSIBILITY AND PLATFORM INTEGRATION

With the rise of mobile technology, crop recommendation systems are
increasingly being developed as mobile apps and web-based dashboards. These
platforms enable farmers in rural areas to access intelligent agricultural tools
without needing advanced computing infrastructure. Flask-based web apps and
Power BI dashboards are commonly used for deployment and visualization.

23
Additionally, cloud computing and IoT integration allow for scalable and real-time
crop recommendations across diverse regions.
3.1.5 INTEGRATION AND USABILITY
Modern systems are designed to be modular and flexible, enabling
integration with soil testing devices, remote sensing platforms, and government
databases. A well-designed user interface ensures ease of use, even for farmers with
minimal technical knowledge. Systems often include language support, mobile
compatibility, and options for downloading reports or sharing recommendations
with agricultural advisors.

3.2 ANALYSIS OF EXISTING METHODOLOGY

3.2.1 DATA COLLECTION METHODS

Crop recommendation systems rely on agricultural datasets collected from
various sources:
• Soil Test Reports: Key parameters such as N, P, K, pH, and moisture content
are collected from field samples.
• Weather APIs: Historical and real-time temperature, rainfall, and humidity
data are sourced from meteorological services.
• Government and Research Datasets: Publicly available agricultural datasets
from institutions like ICAR, FAO, and regional governments are used for
model training.
• User Input: Some systems allow farmers to manually enter their soil and
weather data for localized crop suggestions.

24
3.2.2 DATA PROCESSING AND ANALYSIS
Collected data undergoes several preprocessing steps:
• Data Cleaning and Normalization: Ensures uniform data for accurate model
performance.
• Feature Selection: Algorithms like Recursive Feature Elimination (RFE) are
used to identify the most impactful input features.
• Model Training: Algorithms such as Random Forest, SVM, and Logistic
Regression are trained using labelled datasets.
• Model Evaluation: Accuracy, F1-score, precision, and recall metrics help
assess the effectiveness of the model before deployment.
3.2.3 USER INTERFACE AND EXPERIENCE
Usability is a core concern in existing systems. Key elements include:
• Simple Form Inputs: Soil and weather parameters are entered through
intuitive forms.
• Recommendation Output: Suggested crops are displayed with their
suitability score or rank.
• Visualization: Graphs and charts show trends, nutrient levels, and expected
yield potential.
• Localization: Support for regional languages and units (e.g., Celsius vs.
Fahrenheit) ensures wider adoption.

3.3 CASE STUDIES OR EXAMPLES

3.3.1 RANDOM FOREST-BASED CROP RECOMMENDER

One widely adopted method involves using the Random Forest algorithm to
classify crop suitability. It performs well on agricultural datasets due to its ability

25
to handle nonlinear relationships between input features. These models show high
accuracy in suggesting crops like rice, wheat, maize, or pulses depending on
nutrient composition and weather patterns. Their ensemble nature also helps reduce
the risk of overfitting.
3.3.2 KNN-BASED LOCALIZED CROP PREDICTOR
K-Nearest Neighbours is favoured for systems that require quick adaptation
to local data. By comparing new inputs with nearby data points in the feature space,
KNN provides recommendations that align closely with localized farming
conditions. This method is especially effective when the dataset covers a specific
region or crop cycle.

3.4 CHALLENGES FACED BY EXISTING SYSTEMS

3.4.1 DATA QUALITY AND COMPLETENESS

• Inaccurate or Missing Data: Incomplete or inconsistent soil test data leads to
unreliable predictions.
• Lack of Real-Time Integration: Many systems do not update dynamically
with changing weather or soil conditions.
• Geographical Limitations: Models trained on one region may not generalize
to other regions due to soil and climate variation.
3.4.2 COMPUTATIONAL CONSTRAINTS
• Resource-Intensive Models: Training and deploying high-accuracy models
like XGBoost require significant computing resources.
• Latency Issues: Without optimization, real-time recommendations can be
delayed, affecting usability in fast-paced farming decisions.

26
3.4.3 USABILITY AND ACCESSIBILITY
• Digital Divide: Limited access to smartphones, internet connectivity, or
digital literacy in rural areas restricts system usage.
• Language and Interface Barriers: Lack of regional language support or
complex interfaces reduces farmer engagement.
3.4.4 GENERALIZATION AND MODEL ADAPTABILITY
• Lack of Transfer Learning Capabilities: Many models are trained from
scratch for specific tasks, which limits their ability to generalize knowledge
across different regions or crop types without extensive retraining.
• Insufficient Representation of Diverse Agro-Climatic Zones: Training
datasets may lack diversity in terms of soil types, climate conditions, and
farming practices, leading to poor model performance in underrepresented
regions.
• Dependency on Historical Data Patterns: Models heavily reliant on historical
trends may fail to adapt to sudden changes in climate, pest outbreaks, or
evolving agricultural practices.
• Static Feature Sets: Use of fixed input features can restrict adaptability to
emerging agronomic parameters, such as new soil health indicators or
modern farming technologies.
• Neglect of Socio-Economic Factors: Models often overlook socio-economic
constraints like landholding size, access to resources, and farmer
preferences, which are crucial for generalization and real-world applicability.

27
 CHAPTER 4
PROPOSED SYSTEM
4.1 INTRODUCTION

Crop recommendation is a vital aspect of precision agriculture, aimed at

improving crop yield, reducing resource wastage, and supporting sustainable
farming. Traditionally, farmers relied on intuition or generalized advice to decide
which crops to plant. These approaches often failed to consider key variables such
as soil health, weather conditions, and nutrient levels, leading to reduced
productivity and environmental strain.

With the rise of machine learning (ML) and data analytics, crop
recommendation systems have evolved to become intelligent, data-driven tools that
analyze complex agricultural parameters to provide accurate and timely suggestions.

The proposed system leverages supervised machine learning algorithms to

recommend the most suitable crop based on inputs like nitrogen (N), phosphorus (P),
potassium (K), temperature, humidity, pH, and rainfall. By integrating Python-based
tools like Scikit-learn, NumPy, and Pandas, and deploying the model via Flask for
web accessibility,

The integration of Power BI dashboards enhances data visualization, offering

real-time insights into environmental factors and historical patterns that influence
crop choices. This system is designed to assist farmers, agronomists, and
policymakers in making informed decisions, ultimately contributing to improved
agricultural outcomes and food security.

28
4.2MOTIVATION

The motivation for this project stems from the pressing need to improve crop
planning and farm productivity in the face of climate uncertainty, declining soil
fertility, and resource limitations. Traditional crop selection methods lack the
precision and adaptability needed for today’s farming challenges. Farmers often
experience low yields due to incorrect crop choices, inefficient fertilizer usage, or
unanticipated weather conditions. Additionally, rural farmers may not have access
to expert agronomic advice, leaving them vulnerable to poor planning and financial
losses.

Machine learning offers a reliable, cost-effective solution by analysing historical

and real-time data to make crop predictions tailored to specific environmental
conditions. With increased access to digital tools and soil testing kits, there is an
opportunity to empower farmers with actionable insights that can be delivered via
web or mobile platforms. Applications of crop recommendation systems include:

• Maximizing yield potential by suggesting crops compatible with current soil

conditions.

• Reducing the misuse of resources such as water and fertilizers by aligning

crop needs with available inputs.

This system aims to bridge the knowledge gap by providing intelligent, accessible,
and real-time recommendations that empower farmers with science-backed
decisions.

29
4.3 OBJECTIVES AND GOALS

The proposed crop recommendation system aims to enhance agricultural

productivity and sustainability through data-driven decision-making. The following
objectives guide the system’s design and implementation:

Accurate Crop Recommendation

• Develop a robust model capable of recommending suitable crops based on soil

nutrients and climatic parameters.

• Achieve high accuracy and reliability across different soil types and regional
climates.

• Reduce the risk of crop failure through predictive analytics and optimized
decision support.

Data Preprocessing and Feature Optimization

• Implement data cleaning, normalization, and encoding techniques to ensure

consistent model input.

• Apply feature selection methods to enhance model efficiency and reduce

overfitting.

• Handle missing values and outliers effectively to preserve data integrity.

Scalability and Adaptability

• Design the system to handle large agricultural datasets and adapt to additional
features like crop pricing or pest alerts.

30
 • Ensure compatibility with local, regional, and global datasets to expand
usability.

• Incorporate real-time sensor data and API integrations for dynamic decision-
making.

Interactive Web Interface and Visualization

• Develop a user-friendly interface using Flask, allowing farmers to input

parameters easily and receive instant recommendations.

• Provide real-time visual insights through Power BI dashboards showing crop

trends, soil conditions, and environmental patterns.

• Ensure multilingual and mobile-friendly support for broader accessibility.

Performance and Resource Optimization

• Train and validate models using algorithms like Random Forest, K-Nearest
Neighbors (KNN), and Support Vector Machines (SVM).

• Optimize computational resources to enable fast processing and real-time

feedback.

• Ensure efficient memory usage and potential deployment on lightweight

devices.

Future Enhancements and Continuous Learning

• Integrate advanced models like XGBoost or deep learning techniques (e.g.,

LSTM) for enhanced prediction.

31
 • Explore satellite and IoT data sources for real-time environmental input.

• Enable automatic dataset updates and model retraining for continuous

improvement.

• Strengthen data privacy and security features, especially for cloud-based

deployments.

Integration with Government and Research Databases

• Enable the system to fetch data from agricultural research centres,

meteorological departments, and government databases for enriched,
location-specific recommendations.

• Facilitate alignment with national agricultural schemes and programs to

support broader rural development initiatives.

Educational and Training Support for Farmers

• Include modules or tooltips that explain input parameters and

recommendation results in simple language to assist farmers with limited
technical knowledge.

• Offer downloadable reports or summaries that farmers can share with local
agricultural officers or cooperatives for collaborative planning.

4.4 SYSTEM ARCHITECTURE

The system architecture of the Crop Recommendation System is designed in

modular layers to ensure efficient data processing, accurate prediction, and user-

32
friendly interaction. It starts with the data acquisition layer, which collects real-time
and historical data such as soil type, pH, moisture, weather conditions (temperature,
rainfall, humidity), and crop history.

Figure 4.1 System Architecture

4.4.1 ARCHITECTURE OVERVIEW

The architecture of the Crop Recommendation System comprises several key

components that work cohesively to deliver reliable crop suggestions based on soil
and environmental parameters. These include a data acquisition interface,
preprocessing module, machine learning model (Random Forest, Decision Tree, or
XGBoost), prediction engine, and result visualization module. The system is
designed to be scalable, efficient, and adaptable for both online and offline
deployment, with potential integration of cloud or edge computing for use in remote
agricultural settings.

33
Data Acquisition Module

This module allows users to input essential agricultural parameters such as

Nitrogen, Phosphorus, Potassium, temperature, humidity, pH, and rainfall. It ensures
proper formatting and validation of user input data and supports integration with
sensors or external APIs in future expansions for real-time environmental data.

Data Preprocessing Module

The preprocessing engine handles cleaning and formatting of the input data
before it is passed to the model. Functions include:

• Handling missing or invalid values

• Normalizing values to fit model expectations

• Encoding data where necessary for categorical inputs

• Ensuring consistency across different data sources

4.4.2 CROP PREDICTION MODEL MODULE

This module contains the trained machine learning model (e.g., Random
Forest or XGBoost). It takes the pre-processed data and uses the model to predict
the most suitable crop for cultivation. The model is trained on historical crop yield
data and environmental conditions to ensure accuracy and adaptability to changing
patterns.

34
Recommendation Engine

The engine interprets the output of the machine learning model and converts
it into meaningful insights. It provides the final recommended crop along with
confidence scores. Additional features include:

• Support for ensemble model outputs

• Interpretation of model decision logic

• Scalability for integrating multiple crop categories or regional data

Output and Visualization Module

This module is responsible for presenting the results in a user-friendly way.

It includes:

• A clean and intuitive user interface for web or mobile applications

• Visual display of recommended crops with supplementary environmental insights

• Optional chart or dashboard integration (e.g., Power BI) for deeper analytics

4.4.3 WORKFLOW AND INTERACTION BETWEEN COMPONENTS

All components of the crop recommendation system work in a synchronized

manner to ensure accurate and timely crop suggestions. The process flow is as
follows:

Input and Preprocessing

• Users enter soil and climate-related parameters into the system.

35
• The preprocessing module cleans, formats, and prepares the data for analysis.

Model Prediction and Crop Recommendation

• The machine learning model analyzes the processed input and predicts the most
suitable crop.

• The recommendation engine ensures the result is interpretable and understandable

by the user.

Result Display and Feedback

• The recommended crop is displayed on the screen with visual highlights and
supporting information.

• The voice module announces the crop recommendation for improved accessibility.

• The system may also log the inputs and outputs for future reference or training data
expansion.

Output Generation

• The final count is displayed in real-time on the user interface.

• The detected object count is logged for analysis.

• If enabled, the TTS engine announces the count for accessibility purposes.

36
4.4.4 VISUAL REPRESENTATION

The Visual Representation section illustrates how different modules within the
Crop Recommendation System interact to generate results. It provides insights into
the overall system workflow using visual aids.

Data Flow Diagram (DFD)

Figure 4.2 Data Flow Diagram

A Data Flow Diagram (DFD) visually represents how data moves

through the system:

• User Input: The user enters essential agricultural data such as Nitrogen,
Phosphorus, Potassium, temperature, humidity, pH, and rainfall.

37
 • Data Preprocessing: The entered data is normalized, cleaned, and checked
for validity.
• Model Prediction: The pre-processed data is fed into the trained ML model
(e.g., Random Forest), which predicts the best crop for cultivation.
• Recommendation Engine: The result is interpreted and formatted for user
display.
• Output Generation: The recommended crop is shown on the interface and
optionally announced via voice.
• UI Display: A clean interface shows input fields, predictions, and optional
charts/visuals.
Interaction Flow Chart

Figure 4.3 Flow Chart

A flowchart illustrates the step-by-step processing of input frames:

38
4.5 IMPLEMENTATION DETAILS

The implementation of the Crop Recommendation System is structured to

deliver fast and accurate predictions using machine learning. The system is
developed using Python, with essential libraries like scikit-learn for model training
and Streamlit for building an interactive web interface. Agricultural datasets
containing features such as Nitrogen, Phosphorus, Potassium, temperature, humidity,
pH, and rainfall are used to train models like Random Forest, Decision Tree, and
XGBoost. Input data is pre-processed by scaling and cleaning to ensure consistency
before being fed into the model. Once the prediction is made, the recommended crop
is displayed instantly on the user interface. The system is also equipped with a Text-
to-Speech (TTS) feature to provide voice-based recommendations, enhancing
accessibility for users in rural or visually impaired communities. Performance is
optimized through efficient coding practices and lightweight model deployment,
ensuring real-time responsiveness and ease of use.

4.5.1 TECHNOLOGY STACK

Programming Language

The Crop Recommendation System is implemented using Python, leveraging

libraries such as Scikit-learn, XGBoost, and TensorFlow/PyTorch for machine
learning model development. These frameworks enable efficient model training,
hyperparameter tuning, and seamless deployment.

Machine Learning for Crop Prediction

The system employs supervised learning algorithms such as Random Forest,

XGBoost, and Support Vector Machines (SVM) to predict the most suitable crop

39
based on soil characteristics, environmental conditions, and weather data. These
models are trained on datasets containing features like nitrogen (N), phosphorus (P),
potassium (K) levels, pH, rainfall, and temperature.

NumPy & Pandas

These libraries are used extensively for data manipulation and numerical
operations. Pandas is employed to manage and clean agricultural datasets, while
NumPy powers mathematical operations and array handling for faster computations
during model training

Hardware Acceleration with CUDA

To accelerate training and inference processes, GPU acceleration using

NVIDIA CUDA is integrated with TensorFlow or PyTorch. This improves model
performance, particularly when handling large datasets or deep learning extensions
for image-based classification.

4.5.2 KEY FEATURES

Real-Time Crop Recommendation

The system dynamically suggests crops based on live weather and soil data,
offering timely advice to farmers through a web or mobile interface.

Multi-Factor Prediction System

Using multiple soil parameters and climate indicators, the system

provides crop suggestions tailored to the local agricultural context, ensuring
higher yield and sustainability.

40
Adaptive Learning Mechanism

The model is fine-tuned using feedback from users or agronomists, improving

its accuracy over time through retraining cycles.

Scalability and Dataset Support

The system is capable of supporting various agricultural zones and can be

customized using regional datasets to enhance prediction accuracy based on local
climate and soil conditions.

Visual Dashboard and Data Reporting

Integrated with Power BI or Streamlet, the system presents results in an

intuitive dashboard that includes crop suitability rankings, feature importance charts,
and historical trends.

Multilingual and Voice Support

To increase accessibility for farmers, the system supports voice output in local
languages using text-to-speech engines, and provides a multilingual UI.

4.5.3 DEVELOPMENT METHODOLOGY

Data Collection and Preprocessing

The system uses both open-source agricultural datasets (e.g., Krishi Vigyan Kendra,
ICAR) and custom field-level data. Data preprocessing includes:

• Handling missing values

41
 • Encoding categorical data (like soil type or region)

• Data augmentation for imbalanced crops

Model Selection and Training

Multiple models are trained and evaluated. The best-performing model (e.g.,
Random Forest or Boost) is chosen using:

• Cross-validation

• Hyperparameter tuning (Research)

• Feature importance analysis to eliminate redundant variables

System Integration

The trained model is integrated with a Flask or Django backend, connected to

a database (PostgreSQL/SQLite), and deployed via a web dashboard or mobile app.

Performance Optimization and Testing

The system undergoes model pruning, lightweight deployment using ONNX,

and backend optimization using multi-threading to enhance performance on low-
resource environments.

Deployment and Field Testing

The final system is tested in rural settings or agricultural test beds, evaluating
the recommendation accuracy, usability, and real-time performance on mobile
devices.

42
4.5.4 CHALLENGES FACED DURING DEVELOPMENT

Balancing Accuracy and Simplicity

Designing a model that maintains high accuracy while being explainable and
interpretable for farmers posed a challenge. Simpler models were chosen over
complex ones like deep learning to maintain transparency.

Geographic and Soil Variability

Crop recommendations vary significantly based on local soil and weather

conditions. Gathering region-specific data and ensuring model generalization across
locations required zonal segmentation of training data.

Handling Noisy or Incomplete Data

Field data often had missing or inconsistent entries. Robust preprocessing

methods and data validation checks were used to clean and standardize inputs.

Real-Time Processing and Mobile Support

Running the model on low-end devices and in areas with limited connectivity was
challenging. This was addressed by optimizing the model with TensorFlow Lite and
enabling offline-first functionality.

Crop Overlap and Seasonality:

Some crops share similar growing conditions but differ in seasons. A seasonal
filter was added to the recommendation engine to improve contextual relevance.

43
4.6 TESTING AND EVALUATION

4.6.1 TESTING METHODOLOGIES

During the development of the "Crop Recommendation System using

Machine Learning", a comprehensive testing strategy was adopted to ensure the
system’s accuracy, performance, reliability, and real-world adaptability. Various
testing methodologies were employed to validate each component of the system,
from data ingestion and preprocessing to prediction and visualization.

The primary objective was to verify whether the system could accurately
recommend crops across different soil types, weather patterns, and geographic
zones. Testing was also conducted under variable data quality conditions to assess
model generalization.

Unit Testing

We conducted unit testing to evaluate individual components of the system,

including data preprocessing, feature extraction, and crop prediction algorithms.
Each module was tested independently to ensure correct input-output flow and error
handling. While initial results showed high prediction accuracy with structured
input, further refinements were necessary to handle edge cases such as missing or
inconsistent data from farmers or sensors.

Integration Testing

Integration tests were carried out to validate the seamless interaction between
the data ingestion pipeline, machine learning models, and user interface. Challenges
such as syncing real-time weather data with crop prediction outputs were observed,

44
particularly during network delays. These issues are currently being mitigated
through data caching and asynchronous API requests.

User Acceptance Testing (UAT)

UAT was conducted with a diverse group of users, including farmers,

agricultural officers, and agritech professionals. Their feedback focused on the
accuracy of the crop suggestions, ease of use, and clarity of information presented.
While the majority appreciated the actionable insights and simplicity of the interface,
suggestions were made to enhance local language support and add offline
functionality for use in low-connectivity areas.

4.6.2 EVALUATION CRITERIA

We evaluated the system based on the following criteria, taking into account
both the system was evaluated based on the following key criteria, which were
designed to assess its real-world performance and user impact:

Accuracy of Crop Recommendations

• The system demonstrated high accuracy when predicting suitable crops for
standard soil and climate conditions using models like Random Forest and
XGBoost.
• However, some mismatches were observed in cases involving rare crops or
incomplete environmental data inputs.

System Performance and Efficiency

• The model provided near-instant recommendations for most inputs, ensuring

timely guidance for farmers.

45
 • Occasional latency occurred when retrieving live weather data or working
with larger datasets, which is being addressed through performance tuning and
model optimization.

User Experience and Usability

• Feedback from user testing highlighted the intuitive nature of the interface and
the usefulness of the crop ranking display.
• Suggestions were made to improve visual clarity, provide multilingual
support, and simplify data input for farmers unfamiliar with digital platforms.

4.6.3 RESULTS AND FINDINGS

The evaluation process uncovered both successful outcomes and areas where the
system could be improved:

Successes:

• The machine learning models consistently produced accurate and relevant

crop suggestions across various regions and soil types.
• Users appreciated the simplicity of the system and the inclusion of real-time
environmental data in the prediction logic.

Areas for Improvement

• Prediction accuracy dropped slightly in scenarios involving atypical soil

profiles or unseasonal weather patterns.
• Users requested additional guidance on post-recommendation actions (e.g.,
fertilizer suggestions or pest control strategies) to make the system more
holistic.

46
 • Enhancements are needed to further streamline performance on low-end
devices and under limited internet connectivity conditions

4.6.4 DISCUSSION AND RECOMMENDATIONS

The implementation and testing of the Crop Recommendation System

demonstrated promising outcomes in accurately suggesting suitable crops based on
soil and weather parameters. However, some limitations emerged, such as reduced
prediction accuracy in cases involving underrepresented or region-specific crops.

Introduce dynamic learning: Implement feedback-based updates where farmers

rate the effectiveness of crop recommendations to fine-tune future predictions.

Support for multilingual and audio-based UI: Expand access to semi-literate and
non-English speaking users by integrating voice instructions and local language
translations.

To improve system performance and user satisfaction, several enhancements are

recommended. First, incorporating a feedback loop where users can rate or validate
recommendations will enable dynamic learning and ongoing model improvement.
Second, enabling multilingual support and voice-assisted interaction can make the
system more inclusive, especially for non-English-speaking or semi-literate users.
Offline functionality should also be prioritized for areas with limited internet access.
Furthermore, expanding the system to provide additional guidance—such as
fertilizer usage, irrigation planning, pest alerts, and yield forecasts—would offer a
more comprehensive advisory solution. Lastly, a mobile-first design approach, with
lightweight architecture and simple interfaces, is essential for increasing adoption
among smallholder farmers who primarily use smartphones.

47
 CHAPTER 5

RESULTS AND DISCUSSION

5.1 INTRODUCTION

The Machine Learning-based Crop Recommendation System successfully

identifies the best crop for given soil and environmental conditions. With the use of
a trained machine learning model, the system inspects input features like nitrogen
(N), phosphorus (P), potassium (K), temperature, humidity, pH, and rainfall to
suggest the correct crop. The succeeding sections include observations and analyses
of the system's performance under testing.

5.2 REAL-TIME CROP RECOMMENDATION PROCESS

Overview

The system provides real-time crop recommendations by analysing user-input

data such as soil nutrients (N, P, K), pH, temperature, humidity, and rainfall. Once
the user enters these parameters through a web or mobile interface, the trained
machine learning model (e.g., Random Forest or XGBoost) processes the inputs and
outputs the most suitable crop(s) for cultivation in that specific environment. The
recommended crop is displayed along with a confidence score and additional
insights like optimal sowing time and basic care instructions.

Observations

• The recommendation process is instantaneous and responsive, with results

generated within a few seconds after input submission.

48
 • The system performs well across a variety of input conditions, accurately
suggesting crops such as rice, maize, cotton, or pulses based on soil and
weather features.

• In some cases, when inputs are incomplete or contain extreme/uncommon

values, the system may show lower confidence or suggest broader crop
categories. Continuous model training and real-time data refinement are being
implemented to improve accuracy in such scenarios.

Figure 5.1: Enter the values of quantity of soil

The displayed image shows the user interface of a Crop Recommendation

System using Machine Learning. It allows users to input key agricultural parameters
such as Nitrogen, Phosphorus, Potassium, Temperature, Humidity, pH, and Rainfall.
Based on these inputs, the system predicts the most suitable crop to cultivate using
a trained machine learning model. Once the user fills in the values and clicks the

49
"Get Recommendation" button, the model processes the data and suggests an
appropriate crop, helping farmers make informed decisions for better yield.

Figure 5.2: Recommended crops

The image shows the output interface of the Crop Recommendation System
after the user has entered agricultural inputs and clicked the "Get Recommendation"
button. Based on the values provided for Nitrogen, Phosphorus, Potassium,
Temperature, Humidity, pH, and Rainfall, the system processes the data using a
machine learning model and displays the recommended crop. In this case, the model
suggests that Papaya is the best crop to be cultivated, helping farmers or users make
data-driven decisions for optimal crop selection.

50
 Figure 5.3: Power BI Average Dashboard

This image represents a data visualization from a Crop Recommendation

System dashboard, likely created using Power BI or a similar tool. It shows the
temperature requirement associated with the predicted crop, which in this case is
Rice. The visual highlights that the system takes temperature into account when
predicting the most suitable crop. Rice appears to be recommended based on its
optimal growing conditions aligned with the provided temperature data.

5.3 ACCURACY AND PERFORMANCE ANALYSIS

Overview

The system makes predictions of appropriate crops depending on inputs

provided by the user using an easy-to-use interface. The machine learning algorithm
interprets the data and suggests a crop that is compatible with the soil and climatic
conditions. Every prediction comes with an explanation of why a specific crop was
recommended, enhancing transparency.

51
Observations

• The prediction process is efficient and quick, usually generating results within
less than 2 seconds.
• The system always suggests suitable crops such as rice, maize, sugarcane, and
cotton for their respective optimal conditions.

5.4 PERFORMANCE IN DIFFERENT ENVIRONMENTS

Overview

The Crop Recommendation System was validated with datasets that reflect
various climatic zones, soil types, and seasons to test its flexibility and accuracy
under various agricultural conditions. The performance of the system was
determined in terms of prediction accuracy, consistency, and responsiveness to
changing input parameters.

Observations

• The system works best with full and clean data inputs, especially for weather
conditions and soil nutrients.
• In climatically fluctuating regions or soils with inconsistent profiles, the
system might need tuning to give the best recommendations.
• The model generalizes relatively well to all major crop-producing regions but
can show slightly less precise predictions in regions that are not properly
represented in the training data.

52
 CHAPTER 6

CONCLUSION AND FUTURE WORK

6.1 CONCLUSION

In conclusion, the project "Crop Recommendation System using Machine

Learning" is an intelligent and data-driven system that comes up with a smart and
data-driven approach to assist farmers and agricultural stakeholders to make
intelligent decisions regarding the selection of crops. On the basis of crucial
environmental and soil factors like nitrogen, phosphorus, potassium, temperature,
humidity, pH, and rainfall, the system predicts the best-suited crop for cultivation
accurately. The project is a clear illustration of how technology can close the gap
between conventional agriculture and contemporary smart farming methods,
ultimately encouraging sustainable agricultural practices, enhancing productivity,
and minimizing the risks involved in uninformed crop selection. The ease of use,
scalability, and deployment potential of the system make it a useful tool in the field
of precision agriculture.

6.2 FUTURE WORK

Future enhancements to the Crop Recommendation System using Machine

Learning can focus on improving model accuracy, expanding the variety of crop
suggestions, and increasing overall usability. One key direction is the inclusion of
more diverse, region-specific datasets—such as soil type, rainfall patterns, historical
yield data, and farming practices—which would enable the system to deliver more
localized and precise recommendations. Integration with IoT-based smart agriculture
systems can further strengthen the model by providing real-time sensor data,
53
including soil moisture, temperature, pH, and humidity levels. This dynamic input
would help the system adapt quickly to changing environmental conditions and
provide timely advice to farmers.

In addition to data improvements, the implementation of advanced machine

learning techniques such as deep learning and ensemble models like Random Forest
or Gradient Boosting may enhance prediction performance, especially when
handling complex or high-dimensional datasets. A feedback-driven learning loop can
be introduced where users report the success of recommendations, allowing the
system to refine and adapt over time. Partnerships with agricultural research
institutions and on-ground testing would further validate and strengthen the system
for large-scale deployment in real-world farming environments.

54
 CHAPTER 6

APPENDIX

7.1 APPENDIX 1

PROGRAM CODE

from flask import Flask, request, render_template

import numpy as np

import pickle

import requests

import json

# Load trained model and scaler

model = pickle.load(open('model.pkl', 'rb'))

scaler = pickle.load(open('standscaler.pkl', 'rb'))

# Power BI Push URL (Replace with your actual URL)

POWER_BI_URL = "https://api.powerbi.com/beta/6c94e077-929a-46b2-a642-
8763ee68726c/datasets/a00c252c-b5b7-480b-9ef4-
dc188008a41a/rows?experience=power-
bi&key=TRBH34E5UXsFWoo3oQMANIxRDwWrXmHTv8ZMUk%2BNGdBZq
bfuH%2BEGnEGE7739lmh6zHbFi1M9mXzpa92rlBt1HA%3D%3D"

# Crop dictionary mapping numerical predictions to crop names

55
crop_dict = {

1: "Apple", 2: "Maize", 3: "Jute", 4: "Cotton", 5: "Rice", 6: "Papaya", 7:

"Orange",

8: "Mango", 9: "Muskmelon", 10: "Watermelon", 11: "Grapes", 12: "Coconut",

13: "Banana",

14: "Pomegranate", 15: "Lentil", 16: "Blackgram", 17: "Mungbean", 18:

"Mothbeans",

19: "Pigeonpeas", 20: "Kidneybeans", 21: "Chickpea", 22: "Coffee"

# Initialize Flask app

app = Flask(__name__)

@app.route('/')

def index():

return render_template("index.html")

@app.route("/predict", methods=['POST'])

def predict():

try:

# Collect input values and convert them to float

N = float(request.form['Nitrogen'])

P = float(request.form['Phosporus'])

K = float(request.form['Potassium'])

56
temp = float(request.form['Temperature'])

humidity = float(request.form['Humidity'])

ph = float(request.form['pH'])

rainfall = float(request.form['Rainfall'])

# Create a NumPy array with user input

feature_list = [N, P, K, temp, humidity, ph, rainfall]

single_pred = np.array(feature_list).reshape(1, -1)

# Apply StandardScaler (only one scaler used)

scaled_features = scaler.transform(single_pred)

# Make prediction using trained model

prediction = model.predict(scaled_features)

# Map prediction to crop name

recommended_crop = crop_dict.get(prediction[0], "Unknown Crop")

# Prepare result

result = f"{recommended_crop} is the best crop to be cultivated."

# Send data to Power BI

data = [{

"Nitrogen": N,

"Phosphorus": P,

57
 "Potassium": K,

"Temperature": temp,

"Humidity": humidity,

"pH": ph,

"Rainfall": rainfall,

"Predicted_Crop": recommended_crop

}]

headers = {"Content-Type": "application/json"}

response = requests.post(POWER_BI_URL, headers=headers, json=data)

if response.status_code == 200:

print("Data successfully sent to Power BI")

else:

print(f"Error sending data to Power BI: {response.text}")

except Exception as e:

result = f"Error in prediction: {str(e)}"

return render_template('index.html', result=result)

# Run Flask app

if __name__ == "__main__":

app.run(debug=True)

58
7.2 APPENDIX II – SCREENSHOTS

Figure 7.1: Enter the values of quantity of soil

Figure 7.2: Enter the values of quantity of soil

59
Figure 7.3: Recommended crops

60
 REFERENCES

[1]. Bhagat, P., & Balpande, M., “A comparative study of crop prediction
models using machine learning,” International Journal of Advanced
Research in Computer Science, vol. 9, no. 3, pp. 586–590, 2018.
[2]. Bhattacharya, S., & Samanta, S., “A novel machine learning model
for crop prediction based on climatic parameters,” International Journal
of Innovative Research in Science, Engineering and Technology, vol. 8,
no. 9, pp. 9353–9360, 2019.
[3]. Bendre, M., & Thool, R., “Big data in precision agriculture: Weather
forecasting for future farming,” Procedia Computer Science, vol. 110,
pp. 402–409, 2017.
[4]. Chavan, V., & Bichkar, R. S., “Soil classification and crop suggestion
using machine learning,” International Journal of Engineering Research
& Technology (IJERT), vol. 8, no. 8, 2019.
[5]. Hossain, M. A., & Ghosh, S., “IoT-based smart farming: A machine
learning approach,” International Journal of Advanced Computer
Science and Applications, vol. 10, no. 3, pp. 434–442, 2019.
[6]. Jain, A., & Singh, A., “Crop yield prediction using machine learning,”
International Journal of Scientific Research in Computer Science, vol.
7, no. 6, pp. 42–47, 2019.
[7]. Khaire, S., & Patil, A., “Machine learning based crop recommendation
system,” in Proc. Int. Conf. Intelligent Computing and Control Systems
(ICICCS), 2020, pp. 1236–1240, doi:
10.1109/ICICCS48265.2020.9121210.
[8]. Kumar, R., & Kumari, P., “Crop prediction model using machine
learning approach,” International Journal of Engineering
Development and Research, vol. 7, no. 4, 2019.

61
[9]. Lempitsky, V., & Zisserman, A., “Learning to count objects in
images,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2010, pp.
1324–1332.
[10]. Naik, P., & Patel, D., “Crop recommendation system using decision
tree and random forest,” International Journal of Computer Sciences
and Engineering, vol. 6, no. 10, pp. 243–247, 2018.
[11]. Patil, S. T., et al., “A machine learning approach for smart agriculture,”
in Proc. IEEE Int. Conf. Advances in Computing, Communication and
Control, 2019, pp. 1–5.
[12]. Ramesh, A., & Suruliandi, A., “Soil fertility prediction using machine
learning algorithms: A review,” ICTACT Journal on Soft Computing,
vol. 7, no. 3, pp. 1459–1464, 2017.
[13]. Rani, R., & Tripathi, R., “Prediction of suitable crops using machine
learning,” Journal of Emerging Technologies and Innovative
Research, vol. 6, no. 5, pp. 142–147, 2019.
[14]. Sharma, A., Singh, V., & Shukla, A., “Crop recommendation system

using machine learning approach,” International Journal of Computer

Applications, vol. 179, no. 14, pp. 1–5, 2018, doi:

10.5120/ijca2018917201.

62