Comparing PSO and ACO for XGBoost Heart Rate Classification in IoT Monitoring

Abstract

Advances in Internet of Things (IoT) technology have transformed health monitoring systems, making them more adaptive and capable of acquiring real-time data. Heart rate, measured in beats per minute (BPM), is an important indicator of health status; however, BPM data generated by IoT devices is generally presented as numerical values without automatic classification of normal or abnormal conditions. This study aims to develop an IoT-based heart rate pattern classification system using the Extreme Gradient Boosting (XGBoost) algorithm, optimized with two metaheuristic algorithms Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO), and to compare their performance characteristics. The study employs a quantitative experimental approach, beginning with real-time BPM data acquisition using a MAX30102 sensor and an ESP32-C3 Mini microcontroller from 40 participants, followed by data preprocessing, the development of an XGBoost model, parameter optimization using PSO and ACO, and model performance evaluation. The results show that both algorithms deliver equally good classification performance. PSO successfully identified an optimal threshold range of 68.44–97.82 BPM with an inference complexity of O(1), making it very fast during the prediction stage, while ACO excelled in training convergence speed and feature reduction effectiveness by using only 30% of the total features. The main contribution of this research is a comprehensive comparative analysis of the characteristics of PSO and ACO in terms of computational efficiency, optimization speed, inference complexity, and feature selection effectiveness, resulting in recommendations for selecting the appropriate optimization method to meet the implementation needs of IoT-based health monitoring systems, whether for real-time devices or cloud-based systems.