Measuring Behaviour of People With Dementia Using a Non-Invasive Sensor Network

This paper investigates the application of non-invasive sensor networks and unsupervised machine learning algorithms to measure the behavior of individuals with Mild Cognitive Impairment (MCI) and Dementia (PwD). Utilizing a Lifestyle Monitoring System equipped with door and Passive Infrared (PIR) sensors, this study captures and analyzes movement data to assess Activities of Daily Living (ADLs) within a residential environment. The collected data is processed using a k-means clustering algorithm to categorize behavior into two different classes that indicates whether there are any deviations from the usual pattern. The clustering algorithm achieved a mean Silhouette score of 0.45, indicating a moderate distinction between the categorized behaviors. Additionally, the Pearson correlation coefficients between the clustering results and the predefined labels "no deviations" and "deviation" were 0.4 for breakfast activities and 0.6 for sleeping patterns, supporting the effectiveness of the system in detecting deviations indicative of a progression of the disease. These findings demonstrate that integrating sensor networks with machine learning provides a robust framework for continuous, real-time monitoring, crucial for early intervention and enhancing the quality of life and safety of MCI and PwD individuals.