EO-LGBM-HAR: A novel meta-heuristic hybrid model for human activity recognition

Abstract

Sensor-based Human Activity Recognition (HAR) is widely utilized in various fields, including smart health, security, and smart home applications, to identify a person’s physical activities automatically. One of the main challenges in sensor-based HAR is managing the numerous features collected and extracted from sensor data. As the number of features increases, the complexity of the models also rises, leading to greater demands for computing power and resources—especially in devices with limited power and memory. To tackle this challenge, this study presents a novel hybrid approach called EO-LGBM-HAR, which leverages a metaheuristic-based method to optimize features and enhance the training and evaluation processes of the model, making it more effective and robust. The proposed model combines an Equilibrium Optimizer (EO) as the optimizer with a LightGBM (LGBM) classifier. It has been rigorously tested and compared against the UCI-HAR and WISDM datasets to assess its effectiveness. The results are highly promising, with the proposed model achieving an impressive accuracy of 98.72% on the UCI-HAR dataset and 95.39% on the WISDM dataset.