Assessment of ML Classifiers in Complex Human Activity Recognition Using Wearable Sensors Data

Human Activity Recognition (HAR) is essential for understanding daily behavior patterns, and wearable sensor data serves as a reliable source for monitoring complex activities. This study uniquely evaluates the performance of nine machine learning classifiers in the context of complex human activity recognition, relying solely on wearable sensors. It offers valuable insights into classifier effectiveness for real-world applications. Data from the PAMAP2 dataset, which was collected using three wearable IMUs placed on the hand, chest, and ankle, along with a heart rate sensor, was utilized to identify six daily complex activities. A 70/30 train-test split methodology was implemented to assess classifier performance. The Random Forest (RF) classifier achieved the highest performance, boasting 93% accuracy, precision, recall, and F1-score, followed closely by the K-Nearest Neighbors (KNN) classifier, which recorded 91% across all metrics. In contrast, the Logistic Regression (LR) classifier underperformed, achieving only 55% accuracy, likely due to its limitations in handling non-linear data.These results demonstrate that RF and KNN classifiers are effective for complex human activity recognition, while linear classifiers like LR are less suitable for such tasks. Overall, the Random Forest and KNN classifiers provide reliable performance for complex human activity recognition using wearable sensors, making them excellent choices for practical applications.