Face Recognition Using Machine Learning Models - Comparative Analysis and impact of dimensionality reduction

Abstract

Face Recognition is considered a biometric technique where it is capable of uniquely identifying and verifying a person just by analysing and comparing the facial patterns on the facial contours. Face Recognition has gained significant importance in security aspects and it has been widely used and accepted biometric. It has given greater importance during pandemic situations in terms of cheapest and widely accepted touchless biometrics. This paper studies the impact of dimensionality reduction on the efficiency or accuracy of machine learning algorithms in face recognition. The analysis is carried out over various algorithms include Random Forests, Support Vector Machine, Linear Regression, Logistic Regression, K-Nearest Neighbor. Based on the analysis, Logistic Regression gives better performance in terms of accuracy and time with an accuracy score of 0.97 within a time of 5.74 sec when implemented without principal component analysis whereas with principal component analysis, Logistic Regression achieved an accuracy score of 0.93 within a time of 0. 15sec. There is a huge difference in computation time approximately 20 times, the difference in accuracy is minimal.