Student Name

Date of Submission

Spring 2026

Supervisor

Dr. Shahid Hussain, Associate Professor and Chairperson Computer Science Department, Institute of Business Administration (IBA), Karachi

Committee Member 1

Dr Sajjad Haider, Examiner – I, Institute of Business Administration (IBA), Karachi

Committee Member 2

Dr. Tariq Mahmood, Examiner – II, Institute of Business Administration (IBA), Karachi, Institute of Business Administration (IBA), Karachi

Degree

Master of Science in Data Science

Department

Department of Computer Science

Faculty/ School

School of Mathematics and Computer Science (SMCS)

Keywords

Feature Selection, Genetic Algorithm, Hybrid Feature Selection, High-Dimensional Data, Filter Methods, Wrapper Methods

Abstract

Feature selection is a crucial step in machine learning, especially for high dimensional datasets where the presence of irrelevant and redundant features can negatively impact model performance, and increase computational cost. While filter based methods o!er e”ciency and scalability, they often fail to capture feature interactions. On the other hand, wrapper based methods provide a more flexible search mechanism but are computationally expensive and less stable. This study explores the impact of using filter techniques and Genetic Algorithms, as standalone methods and also as combined hybrids for robust feature selection. A unified experimental framework is proposed to evaluate filter based methods, GA based wrapper methods, and hybrid approaches under consistent conditions. The hybrid approaches include both filters followed by GA and GA guided by filter based fitness functions. Experiments are conducted on four di!erent datasets with varying characteristics, including Gas Sensor, Madelon, Arrhythmia, and Colon Cancer datasets. The methods are evaluated using multiple criteria, including classification accuracy, execution time, feature reduction, and stability. Random Forest is used as the classification model, and cross validation is applied to ensure reliable performance estimation. The results show that filter based methods perform strongly in terms of e”ciency and stability, particularly in datasets where features are individually informative. However, hybrid approaches demonstrate improved performance in complex and high dimensional datasets by e!ectively reducing the search space and enabling better exploration of feature interactions. The GA based wrapper method provides flexibility but incurs higher computational cost. Overall, the study highlights that no single feature selection method is universally optimal, and the choice of method depends on dataset characteristics. Hybrid approaches somewhat o!er a balanced trade o! between e”ciency and performance, making them a relatively better solution for feature selection in high dimensional data.

Document Type

Restricted Access

Submission Type

Thesis

Available for download on Monday, August 16, 2027

Share

COinS