Human Activity Recognition from Biometrics Data Using Kolmogorov-Arnold Network

Human Activity Recognition (HAR) is a feature of an automated system that recognizes human actions. Since most people these days are health-conscious, people use their smartphones or smartwatches to track their daily activities. This helps them organize their schedules and lifestyles more effectively. Recent advancements in Deep Learning (DL) performance have mitigated certain issues related to HAR. Consequently, DL methods are essential for improved competence and precision. This paper provides a comparative study that utilizes state-of-the-art Kolmogorov-Arnold Network (KAN) and Multi-layer Perceptron (MLP) to classify human activities using biometrics data. The Biometrics dataset, which includes 18 classes representing a variety of activities, is used for HAR. For optimal outcomes, the suggested algorithm is trained and tested using the TensorFlow structure and a hyperparameter tuning technique. The outcomes show that the KAN algorithm performs quite well in identifying human activity with an accuracy of 72.64% and a loss rate of 0.9136. The experiment's findings suggested that the KAN model performs more effectively and accurately for human activity identification.