Stack-HAR: Complex Human Activity Recognition With Stacking-Based Ensemble Learning Framework

Human activity recognition (HAR) is crucial in various domains, such as healthcare, elderly care, sports, gait analysis, and security surveillance. Despite its critical role in these fields, achieving high accuracy in HAR remains a challenging task. This article proposed a multimodel stacking-based ensemble learning framework for complex HAR (Stack-HAR). The proposed framework demonstrates significant performance improvements across benchmark datasets, such as mHealth, UCI HAR, and WISDM datasets. By leveraging the stacking-based ensemble technique, the proposed Stack-HAR framework achieves exceptional accuracy of 99.49%, 96%, and 90% on these datasets, respectively. The metalearner consistently outperforms individual classifiers, showcasing the effectiveness of ensemble learning. This framework offers a balanced solution for real-world HAR applications, optimizing both accuracy and computational efficiency. Recognizing activities of daily living (ADL) using smartphones, given their widespread sensor availability, gets benefited greatly from this approach.