Visualization Driven AI Ensuring Quality in Wearable Time-Series Sensor Data for Healthcare

Human Activity Recognition (HAR) plays a critical role in wearable sensor systems and health monitoring. While machine learning models are used for activity classification, visualization techniques have a great deal to provide in model performance improvement, feature selection, and interpretability. This paper examines the impact of visualization techniques on HAR data and demonstrates how visualization can contribute towards improving classification accuracy. Experimental results confirm that feature engineering using visualization-based techniques considerably improves model performance. We assess several visualization approaches, their role in HAR systems, and how they are applied in real-time settings. We also describe how visualization facilitates the interpretation of sensor signals, minimization of misclassification errors, and the extraction of informative activity patterns, which results in robust and interpretable HAR models.