Performance of Face Recognition Machine Learning Algorithms in Attendance Recording System with Limited Training Data

The effectiveness of the face recognition algorithm is crucial to a digital attendance system. Differences in lighting, camera, room size, and the number of students create a challenging environment for face detection, but one problem that is rarely discussed in face recognition is the problem of limited training data. In attendance recording scenarios, attaining a large number of participant image data might not be feasible due to time constraints or other limitations. The research comprises several key stages, including application requirements analysis, development of the Attendance System application, data collection, development of artificial intelligence (AI) components, and application testing. Within this framework, the study compares the performance of three face-recognition machine learning algorithms (SVM, KNN, and Random Forest) under limited training data conditions. It is demonstrated that the best performing algorithm (SVM with hyperparameter tuning) resulted in the highest accuracy of 0.45. When the data quality is increased by removing some anomalies, the algorithm performs at a higher accuracy of 0.61. The effect of limited data on training algorithms is then examined and discussed further.