Accuracy Comparison of Support Vector Machine and K-Nearest Neighbors in Face Recognition for Library User Identification

Abstract

Traditional library book lending systems that rely on membership cards or personal IDs are prone to misuse due to human error. To address this, this study developed a web-based book lending application using face recognition enabling automatic user verification without physical cards, improving security, and reducing human errors. In this research 10 university students took roles as the application’s users. The goal is that the application is able to identify every library user who is going to borrow or return books based on their real time face image. The face recognition itself has been developed using dlib’s face detection, cropping, and feature extraction functions and Support Vector Machine (SVM) classification model. The K-Nearest Neighbors (KNN) model was also tested to for classification accuracy comparison. Model validation tests show that the dlib works well in detecting face location within an image, cropping the face area, and extracting face features while the two classification models are able to well classify student IDs too. The SVM model results in 91% accuracy, 90% precision, 91% recall, and 91% F1-score, which is however slightly better than KNN’s 89% accuracy, 89% precision, 88% recall and 88% F1-score. The SVM has been then chosen for the application. Following the completion of application development, a system test has been conducted with black box method and returns with system accuracy of 90%. This finding confirms that implementing dlib and an SVM model for user identification for an application can be a promising method.