Case Studies: Daily Activity Monitoring

Md Zia Uddin

Abstract: This chapter focuses on practical, Python-based guidelines for analyzing human behavior and health data using wearable and ambient sensors, illustrating how deep learning and explainable artificial intelligence can support independent living and healthcare monitoring. The core motivation behind these methods lies in the limitations associated with wearable sensors—specifically, discomfort during prolonged use and potential rejection by elderly individuals who may find them cumbersome. Conversely, ambient or external sensors typically achieve higher acceptance in real-world settings, provided they reliably capture accurate data from a distance. The chapter brings attention to these trade-offs, introducing multiple strategies to handle challenges such as user compliance, data variability, and complex sensor installation procedures. Several publicly available datasets are explored, alongside actual real-time examples, to demonstrate the feasibility of implementing deep learning algorithms for behavior recognition. These datasets further validate the adaptability of Python-centric approaches, emphasizing how simple or advanced models (e.g., decision trees, random forests, support vector machines, and neural networks) can be implemented, optimized, and evaluated for real-time predictions. Code snippets and graphical outputs illustrate step-by-step methodology, ensuring readers can replicate or extend the experiments with minimal setup. The chapter also spotlights the Opportunity dataset, a richly labeled resource from a smart home environment that integrates signals from both body-worn sensors and ambient devices, reinforcing the importance of context in activity detection. Throughout, the narrative underscores the significance of explainable AI in transparent decision-making. Techniques like Local Interpretable Model-Agnostic Explanations (LIME) are showcased to interpret classifier outputs and uncover which features drive specific predictions. This transparency becomes vital for user trust, especially in healthcare scenarios where model recommendations may inform critical interventions or lifestyle adjustments. Ultimately, the chapter offers a comprehensive overview of sensor-based data collection, pre-processing pipelines, and modeling solutions under the umbrella of machine learning and explainable methods.

External IDs:doi:10.1007/978-3-031-97359-8_5