1

Analytics and Decision Science

Student Name

Institution Affiliation

Date
 2

Part 1

Introduction

Analytics refers to computational analysis of data or statistics used to discover

information and communicate meaningful patterns in a given set of data. The process is also

meaningful in helping organizations making effective decisions regarding various datasets

(Davenport, 2013). The process involves using various statistical methods to analyze the data to

gain various insights and make relevant decisions. On the other hand, decision science uses

quantitative techniques to inform the making of decisions in various organizations. The decision

scientists analyze the given data and use various decision-making methods such as Analytical

Hierarchical Process (AHP) and TOPSIS in arriving at various decisions. The decision scientists

apply mathematics, business, and technology, among other disciplines to transform the data and

make relevant decisions (Roy, 1993). The paper herein will focus on helping the manufacturing

company herein to select the best way to acquire more space. The company has four alternative

options of either extending the current warehouse, purchasing space in either the city or suburb

based on the five criteria, including commuting, client parking, warehouse space, and security of

the company and its employees. The paper herein will use Analytical Hierarchy Process (AHP)

and TOPSIS method to model the decision for the company. After the analysis, a response will

be provided to the company advising them on the way forward and the best alternative to acquire

more space for the company based on the analysis made. It will also provide more information,

including technical information on how the decision was achieved and how the solution was

obtained.

TOPSIS ANALYSIS
 3

TOPSIS analysis is one of the methods used in the making or helps organizations arrive

at various decisions. It's a Multi-criterion decision analysis method that is widely used by

decision scientists to make relevant decisions since its simple, rational, and also has a good

computational efficiency which measures the performance of each given alternative and helps

the decision scientists to arrive at a decision (Balcerzak & Bernard, 2017). The analytical method

was developed in the early 1980s and has undergone various improvements over the years. The

decision analysis method is based on the Positive Ideal Solution (PIS) and the Negative Ideal

Solution (NIS) to select the best alternative. The alternative to be selected needs to have the

shortest geometric distance from the P and the longest geometric distance from the NIS. The

method compares the provided alternatives and uses the company's criteria to identify the best

viable option or alternative that the company can take to acquire more space in the case herein.

The method herein starts with creating an evaluation matrix where the values on a 5-point scale

replace the descriptive values. The evaluation matrix is then normalized. The criteria weights and

the best and worst alternatives are calculated as well as the Euclidean distance between the target

alternative and the worst condition or alternative. The last step of the method herein involves

ranking the given alternatives to identify the best and the worst options or alternatives that the

organization or the company can take in the given case (Chen, 2019). The 5-point scale of

relative importance used in the case herein is as follows.

1-Unimportant

2-Slightly Important

3-Moderately Important

4-Important

5-Very Important
 4

Table 1: Evaluation Matrix

Evaluation Matrix
Commuting Client Warehouse Security Cost
parking Space
A1 (Center) 5 1 5 2 800,000
A2 (Suburb) 1 4 3 3 600,000
A3 (Shared) 5 5 1 3 150,000
A4 (Extend) 2 2 2 3 200,000
Evaluation 7.416198487 6.78232998 6.244997998 5.56776436 1030776.40
matrix value 3 3 6
The obtained evaluation matrix is standardized and multiplied by the weighted value of

the criteria, 0.20, to create a normalized weighted matrix. The ideal value is the worst. The bet is

then identified from the normalized weighted matrix and then used in the Euclidean distance

calculation, which is then used in the performance score calculation that ranks the given

alternatives identifying the best alternative company can take to solve its problem herein.

Table 2: Normalized Weighted Matrix

Normalized
Weighted Matrix
Commu Client Wareho Securit Cost
Euclide Euclide Perfor Ra
ting parking use y an an mance nk
Space Distanc Distanc Score
e (ideal e (ideal
best) worst)
A1 0.13483 0.02948 0.16012 0.07184 0.15522 0.21120 0.14166 0.40146 2
(Cent 9972 8391 8154 2121 28 0231 4295 9359
er)
A2 0.02696 0.11795 0.09607 0.10776 0.11641 0.18594 0.22144 0.54357 1
(Sub 7994 3565 6892 3181 71 2996 2837 0293
urb)
A3 0.13483 0.14744 0.03202 0.10776 0.02910 0.23109 0.23109 0.5 3
(Shar 9972 1956 5631 3181 4275 8576 8576
ed)
A4 0.05393 0.05897 0.06405 0.10776 0.03880 0.15365 0.15365 0.5 3
(Exte 5989 6782 1262 3181 57 1233 1233
nd)
 5

The normalized weighted matrix above shows that alternative number 2 is the most

viable and suitable option for the manufacturing company to expand or acquire more space since

it's the alternative with a high-performance score. Alternative 2 has a performance score of 50%,

followed by alternative 1 with a performance score of 40%, whereas alternative 3 and 4 tie with a

performance score of 5% each.

ANALYTIC HIERARCHY PROCESS

The AHP method is a decision-making process that uses a structured way in modeling the

decision for the task or problem at hand. The process of the method makes use of mathematics

and psychology in organizing and making relevant decisions for the companies. It's a decision-

making method developed in the 1970s and has undergone various improvements ever since

(Saaty, 2008). It comprises three parts: the first part, which is the goal to be achieved, the

available alternatives, and the criteria that help in selecting the available alternatives, all of which

are given for the small manufacturing company herein. The multi-criterion decision-making

method herein helps in making relevant decisions for the company to select the best decisions,

using the available information criteria weights for the various alternatives available for

consideration of the company to provide the best alternative for the company based on the

percentage weights of the alternatives obtained from the analysis of the available data (Cheng &

Li, 2001). The method involves various processes or methods first, including the creation of the

hierarchical structure where the goal, criteria, and alternatives take the first, second, and third-

level, respectively.
 6

Figure 1: Hierarchical Structure

The second step of the method herein is to create a pairwise matrix done using the scale

of relative importance. The step involves identifying the importance of the criteria available

concerning the company's goal herein, acquiring more space. The scale of relative importance to

be used in the case herein is as follows.

1-Unimportant

2-Slightly Important

3-Moderately Important

4-Important

5-Very Important

The second step of the decision-making method herein is creating the pairwise matrix,

which determines the relative importance of the various criteria used herein or considered by the

company to decide or select an alternative.

The normalized pairwise matrix is then created from the pairwise matrix shown above by

dividing the values in every column by its sum. The normalized pairwise matrix, criteria weights

for different criteria considered by the company herein, are calculated in the figure below. The

consistency of the values used in the matrix is also calculated to determine the accuracy of the

matrix and viability of the matrix to be used in further decision making. The consistency ratio of

0.058 is obtained, which means that the matrix is consistent and can be used to further the

company's decision-making processes (Alonso et al., 2006).

Figure 2: Normalized Pairwise Matrix

The figure above provides the criteria weights that the company herein needs to give to

the various criteria that the company will consider in achieving its goal herein, acquiring more

space. The criterion that should be considered first has high criteria weights, and for the case

herein, they include commuting, client parking, warehouse space, security, and cost in the given

order.

Response to the Company

The small manufacturing company herein needs to adopt the Analytic Hierarchy Process

(AHP) analysis to acquire more space or achieve its goal. The method is the most viable method

for helping the company select the best option or alternative in meeting its need, acquiring more

space. It follows well-researched and developed procedures where the relation between the

criteria used in the selection process is determined. The criteria weights for the criteria are

calculated; thus, the method provides efficient and accurate results that help make decisions. The

AHP method first provides a hierarchical structure for the decision problem for the company

herein, which helps the decision scientist to understand the goal that the company wishes to

achieve, unlike the TOPSIS method. From the hierarchical structure, a pairwise matrix is

calculated based on the relation of the given criteria towards the goal to be achieved and then

standardized to form a normalized pairwise matrix used in the calculation of the criteria weights.

The consistency level of the matrix is then checked using the constructed matrix to determine the

viability of the matrix and decide whether it can be used for further decision making of which in

the case herein a consistency level of 0.058, which shows that it is viable and good for further

decision making is obtained. The criteria weights are then calculated and ranked to identify the

best alternative to solve the company problem herein. With the information on how important
 9

the given criteria are, the company can make the relevant decision and choose the best alternative

available based on the criteria.

Reflection

The company should use the AHP method since it's a simple method that helps an

organization make viable decisions. It individually calculates the importance of the various

criteria that the company needs tso consider in making the decision, thus enabling the decision-

makers to make relevant decisions since they know exactly what is needed and what is not

required to select the best alternative to accomplish the company goal of acquiring more space.

The TOPSIS method presents certain drawbacks, such as rank reversal, which causes the given

alternatives to change when an alternative is added or removed from the decision problem. The

disadvantage herein can lead to a total rank reversal where the alternative considered the best can

become the worst whereas the worst alternative can be the best. Thus, this reason makes it not a

suitable decision method for the company herein. The TOPSIS method also has a high

subjectivity, making it prone to errors and thus not suitable for providing a practicable solution

for the problem herein.

Part 2

Introduction

Missing data is a usual issue in data science and analysis, which occurs in almost every

analysis process. It's the value that is not taken or recorded during data collection and can have

variant effects on the reliability and viability of the collected data. Most data analysis procedures

cannot be performed when a dataset has missing values, and thus the value needs to be dropped

or ignored, or it needs to be filled (Maniruzzaman et al., 2018). There exist various ways in
 10

which missing values are filled in statistics though the most common one is through calculating

the mean and replacing the missing value with the mean of the column where it's located. The

second part of this assignment will determine how missing values are imputed, tackling the

various approaches, using a clustering algorithm to cluster the dataset provided hierarchically, or

using a hierarchical clustering algorithm and explaining the dendrogram provided.

Imputation

Imputation is the process where missing values are substituted with other values in a

dataset (Madley-Dowd et al., 2019). The imputation process involves identifying the missing

values in a dataset and replacing them with a numeric value that can be calculated using various

approaches, some of which are discussed below.

Using Mean/Median Values

Many data scientists and analysts most often use the imputation using medium or median

values to impute missing data. The method is simple to use and results in good algorithm

performance, and therefore provides a good overview of the data. The mean and median of the

non-discarded values are obtained using the mean or median value of the column in each column

where the missing values are replaced. It's a commonly used method in numerical data, and it

provides accurate results with smaller datasets. However, the method discussed herein fails to

account for the various uncertainties in the imputed data and sometimes gives poor results.

Imputation Using Multivariate Imputation by Chained Equation (MICE)

This method of imputation works by repeatedly filling in the blanks. MIs are far superior

to single imputations in measuring the uncertainty of missing values (Putri et al., 2018). The

chained equations method is also very versatile. It can handle a wide range of variables and data

types (continuous or binary) and complications like bounds and survey skip patterns.
 11

Using most Frequent or Zero Constant values

Imputation using the most frequent values is a mathematical technique for imputing

missing values that is most often applied to categorical data or categorical features. The method

is commonly used for this type of data, and it replaces missing values with the most common

values in the column. On the other hand, the process ignores the association between the features

and can result in dataset bias.

For the Brew Dog dataset provided herein, the SPEED BIRD beer is the one missing its

price. The price of the beer is not included in the dataset and thus needs to be substituted with a

value before continuing with the next operation. The imputation approach or method used herein

is the use of mean where the mean of the price column of the dataset will be calculated and

substituted for the value of SPEED BIRD beer. The imputation approach or method is chosen

because it’s easy to use and provides accurate results that help the analyst obtain accurate

insights from the data. Excel will be used in calculating the mean of the price column where the

individual prices of different beers will be summed and then divided by the total number of

entries which is 15 in this case. The mean obtained will be the new price of the SPEED BIRD

beer and will be used to fill the blank space. As shown in the figure below, the obtained mean is

4.92, and thus the blank space of SPEED BIRD beer in the price column will be replaced by the

mean, which is 4.92.

Figure 3:BrewDog Dataset with Mean

Clustering

Clustering is the division or separation of data points into groups of similar

characteristics or features. It's an unsupervised learning approach for identifying and grouping

data points in large datasets without regard for a specific result. Since data is typically clustered

based on certain attributes, it aids in grouping data into easily understandable groups or clusters

(Rodriguez et al., 2019). The following are the two most common clustering algorithms that are

commonly used.
 13

K-Means Clustering

K-means clustering algorithm is a commonly used algorithm that iteratively finds the

local maxima in each iteration of its steps. When using the algorithm, the analyst has first to

identify the algorithm and specify the number of clusters that the algorithm needs to group or

cluster. The algorithm computes the centroids and iterates the results until the optimal centroid is

found. It's an unsupervised machine learning algorithm that is used with unlabeled data. Data

points are grouped based on their similarity in features.

Hierarchical Clustering

Hierarchical clustering is a clustering algorithm mostly used in clustering and forms

clusters in a hierarchical way. It is a cluster analysis approach that aims to establish a hierarchy

of clusters. The two types of hierarchical clustering include Agglomerative hierarchical

clustering, a bottom-up approach where the clusters are built from single clusters that are merged

upwards as the hierarchy goes up (Murtagh & Contreras, 2017). Another type of hierarchical

clustering is divisive hierarchical clustering. In this top-down approach, observations begin in a

single cluster and split into different groups as one moves down the cluster. The clustering

algorithm displays its results using a dendrogram which shows the relationship between the

objects and the arrangement of the objects (Cohen-Addad et al., 2019). The R programming

language will perform hierarchical clustering for the assignment herein and provide the

dendrogram as the output. The Brew Dog dataset will be loaded or imported in the R studio, and

a code written to perform hierarchical clustering is written. From the results obtained, the dataset

is first grouped or divided into two clusters or groups. The first group does not subdivide and

goes into the last phase listing the individual beers, including speed bird, Hazy Af, and lost Lage.
 14

The second group is further subdivided into two, where one group is divided into two subdivided

ends, whereas the other is further subdivided into two.

Figure 4: Dendrogram

Reflection

The clustering technique carried out herein; hierarchical clustering is a powerful and

widely used clustering technique that uses the dendrogram to display the relationship between

the different objects in a given dataset. It provides an in-depth analysis of how the different types

of beers that the company deals with relating to the other using the dendrogram, which is easy to

interpret and easy to create. Unlike the other clustering technique, a hierarchical clustering

output, a dendrogram, can be easily explained or interpreted by a person with no statistical

background since it's easy to understand and implement. However, the method has its own

limitations, including failure to work with missing values or data, unlike the K-means clustering,

which works even with missing values. The method does not provide the best solution since it

involves a lot of arbitrary decisions.

Reference List

Alonso, J.A. and Lamata, M.T., 2006. Consistency in the analytic hierarchy process: a new

approach. International journal of uncertainty, fuzziness and knowledge-based systems, 14(04),

pp.445-459.

Balcerzak, P. and Bernard, M.P., 2017. Digital economy in Visegrad countries. Multiple-criteria

decision analysis at regional level in the years 2012 and 2015. Journal of Competitiveness, 9(2).

Chen, P., 2019. Effects of normalization on the entropy-based TOPSIS method. Expert Systems

with Applications, 136, pp.33-41.

Cheng, E.W. and Li, H., 2001. Analytic hierarchy process. Measuring business excellence.

Cohen-Addad, V., Kanade, V., Mallmann-Trenn, F. and Mathieu, C., 2019. Hierarchical

clustering: Objective functions and algorithms. Journal of the ACM (JACM), 66(4), pp.1-42.

Davenport, T.H., 2013. Analytics 3.0. Harvard business review, 91(12), pp.64-72.

Madley-Dowd, P., Hughes, R., Tilling, K. and Heron, J., 2019. The proportion of missing data

should not be used to guide decisions on multiple imputation. Journal of clinical

epidemiology, 110, pp.63-73.
 16

Maniruzzaman, M., Rahman, M.J., Al-MehediHasan, M., Suri, H.S., Abedin, M.M., El-Baz, A.

and Suri, J.S., 2018. Accurate diabetes risk stratification using machine learning: role of missing

value and outliers. Journal of medical systems, 42(5), pp.1-17.

Murtagh, F. and Contreras, P., 2017. Algorithms for hierarchical clustering: an overview,

II. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 7(6), p.e1219.

Putri, B.D., Notobroto, H.B. and Wibowo, A., 2018. Comparison of MICE and Regression

Imputation for Handling Missing Data. Health Notions, 2(2), pp.183-186.

Rodriguez, M.Z., Comin, C.H., Casanova, D., Bruno, O.M., Amancio, D.R., Costa, L.D.F. and

Rodrigues, F.A., 2019. Clustering algorithms: A comparative approach. PloS one, 14(1),

p.e0210236.

Roy, B., 1993. Decision science or decision-aid science?. European journal of operational

research, 66(2), pp.184-203.

Saaty, T.L., 2008. Decision making with the analytic hierarchy process. International journal of

services sciences, 1(1), pp.83-98.