Research Article | Open Access | Download Full Text
Volume 4 | Issue 1 | Year 2025 | Article Id: DST-V4I1P104 DOI: https://doi.org/10.59232/DST-V4I1P104
Comparative Study of Machine Learning Algorithms for Facial Recognition Systems
Arun Kumar Singh
| Received | Revised | Accepted | Published |
|---|---|---|---|
| 11 Jan 2025 | 12 Feb 2025 | 16 Mar 2025 | 31 Mar 2025 |
Citation
Arun Kumar Singh. “Comparative Study of Machine Learning Algorithms for Facial Recognition Systems.” DS Journal of Digital Science and Technology, vol. 4, no. 1, pp. 50-62, 2025.
Abstract
Robust facial recognition systems are frequently adopted in security, healthcare and entertainment fields and improve substantial growth through ML algorithms. This paper compares different machine learning approaches: CNN, SVM, k-NN, Decision Trees and others with a focus on the facial recognition task. The algorithm is evaluated based on accuracy, speed, efficiency, and insensitivity to changes in lighting and pose of the subject. In order to avoid bias when evaluating the performance of the developed algorithms, the study is performed on recognized datasets like Labeled Faces in the Wild (LFW) and Yale Face Database B. Outcome shows that DL architecture based on CNN yields superior performance than other traditional ML but at a higher computational load. The paper’s conclusion is presented through recommendations on how best to select an algorithm given an application and outlined future studies. Biometrics and facial recognition have been receiving considerable attention worldwide in recent years due to improvements in ML that improve their accuracy, performance, and flexibility of use. In this context, this work aims to compare well-known techniques of facial recognition based on ML algorithms using methods like CNN, SVM, PCA, and DBN. The performance of each algorithm is compared based on accuracy, time of execution, consistency under varying illumination and computational complexity. This analysis shows that although deploying CNN-based architectures always provides high accuracy and flexibility, it requires significant computational power that cannot practically be used for low-power devices. On the other hand, algorithms such as PCA have lower computational time in recognizing, but they are not efficient when dealing with high dimensional data, which also affects the recognition rate under complex circumstances. In this paper, various approaches that balance the accuracy and computational efficiency of the algorithm are presented as conclusions for choosing the best algorithm depending on the application. The insights from the work will benefit the exploration of future directions and the practical applicability of facial recognition algorithms by focusing on context-specific algorithm choices.
Keywords
Facial Recognition, Machine Learning, Convolutional Neural Networks (CNN), Support Vector Machines (SVM), Principal Component Analysis (PCA), Deep Belief Networks (DBN), Algorithm comparison, Biometric authentication.