Leveraging Ensemble Learning Models for Human Activity Recognition

This paper presents a novel method for categorizing human activities by processing sensor data obtained from IoT devices, focusing on improving accuracy. The proposed approach leverages an ensemble learning framework with majority voting, integrating hyperparameter-optimized classifiers to enhance predictive performance.The ensemble approach minimizes individual biases and errors, effectively handling the variability inherent in sensor data. Adequate preprocessing techniques refine data quality before feeding it into the model. A diverse set of base classifiers, such as KNN, Decision Tree, and Random Forests, are considered for classification. Hyperparameter-optimized KNN Grid Search, Gradient-Boosted Decision Trees, and Random Forests with Optimal Trees are ensembled. Extensive experiments were conducted on Human Activity Recognition datasets, WISDM, HAPT, HAR, and KU-HAR.The model performance was rigorously evaluated using classification metrics such as accuracy, precision, recall, and F1-score. Empirical results demonstrate that the proposed ensemble method significantly enhances classification accuracy. Future research will investigate applying deep learning techniques to capture complex feature interactions within sensor data better.