Analysis of Machine Learning Models for Wearable Devices Centric Human Activity Recognition

This work presents a comprehensive study on applying machine learning models for human activity recognition. The increasing number of wearable devices and the growing interest in health monitoring and intelligent environments have fueled the demand for robust and accurate systems that automatically recognize human activities. This study explores and compares the performance of different machine learning (ML) models in human activity recognition. This study intends to identify the most effective ML model to classify and predict human activities accurately. In this study, various models are evaluated, including ensemble-based classifiers such as Decision Tree, Random Forest, K-Nearest Neighbor, Logistic Regression, AdaBoost algorithm, and XGBoost algorithm. All models have been evaluated using the publicly available dataset “Human Activity Recognition with Smartphones,” which captures six daily activities: Lying, Standing, Sitting, Walking, Walking Upstairs, and Walking Downstairs. Among all the applied models, XGBoost accomplished the highest accuracy of 99.52% with 100% precision, 100% recall (100%), and 1.0 F1 score. Hyperparameter tuning on the ML models is implemented to attain the best accuracy.