KHAJA BANDANAWAZ UNIVERSITY,KALABURAGI
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
TECHNICAL SEMINAR
“Machine Learning for Anomaly
Detection”
UNDER THE GUIDANCE OF: PRESENTED
NAME: KASHINATH .S. MUDHOL
PROF. mrs ayesha kiran
UIN : 21KB02BS[027]
preencoded.png
CONTENTS
1.INTRUDUCTION
2.SUPERVISED LEARNING METHODS FOR ANOMALY DETECTION
3. UN SUPERVISED LEARNING METHODS FOR ANOMALY DETECTION
4. SEMI- .SUPERVISED LEARNING METHODS FOR ANOMALY DETECTION
5. EVALUTION METRICS FOR ANOMALY DETECTION MODELS
6.REAL-WORLD APPLICATIONS OF ANOMALY DETECTION
7. CHALLENGES AND LIMITAIONS OF ANOMALY DETECTION
8. FUTURE TRENDS IN AMOMALY DETECTION
9. CONCLUSION
preencoded.png
Introduction :
Anomalies are data points that established from normal behavior. Examples
include fraud, defects, intrusions, and medical outliers.
Anomaly detection is important in cybersecurity, finance, healthcare, and
manufacturing. Machine learning methods are adaptive and detect subtle
anomalies.
preencoded.png
Supervised Learning Methods for Anomaly
Detection
Supervised Learning Advantages
Meaning each data point has a corresponding correct output • High accuracy when labeled data is available
to predict outcomes or make classification on new, unseen • Direct control over the types of anomalies detected
data .
Limitations
• Requires large labeled datasets
• Performance degrades with unseen anomaly types
preencoded.png
Unsupervised Learning Methods for Anomaly
Detection
Unsupervised Learning Advantages
unsupervised learning deals with datasets where data points • No need for labeled data
don't have pre-assigned labels. • Can detect abnormal anomalies
Limitations
• Lower accuracy than supervised methods
• Sensitive to parameter tuning
preencoded.png
Semi-Supervised Learning
Methods for Anomaly
Detection
One-Class SVM
Gaussian Mixture Models (GMM)
Deep Support Vector Data Description (Deep SVDD)
Train on normal data and detect deviations. Combines supervised and
unsupervised benefits. Requires less labeled data, robust to noise, and
effective for subtle anomalies.
preencoded.png
Evaluation Metrics for
Anomaly Detection Models
Precision Recall
Proportion of correctly Proportion of actual anomalies
identified anomalies. detected.
F1-score
Harmonic mean of precision and recall.
Imbalanced datasets and varying anomaly types pose evaluation challenges.
Cost-sensitive evaluation considers false positives and negatives.
preencoded.png
Real-World Applications of
Anomaly Detection
Fraud Detection in Finance
Intrusion Detection in Cybersecurity
Predictive Maintenance in Manufacturing
Healthcare Anomaly Detection
Applications in environmental monitoring, autonomous driving, and IoT security.
Anomaly detection is extremely versatile!
preencoded.png
Challenges and Limitations of Anomaly Detection
Imbalanced datasets
1
2 Concept drift
3 Lack of labeled data
Techniques like oversampling, adaptive learning, and unsupervised methods can address these challenges.
preencoded.png
Future Trends in Anomaly
Detection
Explainable AI (XAI)
Federated learning
Graph-based detection
Automated Machine Learning (AutoML)
preencoded.png
CONCLUSION
This systematic literature review studied anomaly detection through machine learning techniques (ML). It
reviewed ML models from four perspectives: the application of anomaly detection type, the type of ML technique,
the ML model accuracy estimation, and the type of anomaly detection (supervised, semi-supervised, and
unsupervised). The review investigated the relevant studies that were published from 2000-2020. We queried 290
research articles that answered the four research questions (RQs) raised in this review.
preencoded.png
preencoded.png
