Enhancing Human Movement Classification in Imbalanced Data with a Hybrid AI Approach

Abstract

The monitoring and classification of human movements have garnered significant importance across various domains, including hazardous situation detection and elderly care. In recent years, the application of Artificial Intelligence (AI) has emerged as a promising approach to predicting and analyzing human movements in such contexts. However, as the complexity and diversity of target movements increase, obtaining sufficient and high-quality data for AI methods becomes increasingly challenging. A particular obstacle researchers and practitioners encounter is the uneven distribution of movement data, where certain movements are frequently repeated and well-represented in the dataset, while others occur infrequently, leading to an imbalanced dataset. This data imbalance poses a critical issue as it hampers the effectiveness of AI models, particularly for those movements with limited available data. The scarcity of samples for certain classes can lead to biased learning, resulting in reduced accuracy and poor generalization of the model. To address this critical challenge, this research proposes a hybrid AI methodology specifically designed to enhance the classification accuracy of two important human movements: turning and walking. The experimental results demonstrate the efficacy of the proposed method. It significantly improves the overall classification accuracy compared to only one AI technique, especially for those movements characterized by limited data availability. By effectively addressing the data imbalance problem, the hybrid AI methodology ensures that the model's performance is not disproportionately skewed towards the overrepresented classes, yielding a more balanced and reliable human movement monitoring and classification system.