Abstract:
In this paper, we present a rule-based approach to the inference of elders’ activity in two primary application areas: detecting Independent Activities of Daily Living (IADLs) for the detection of anomalies in activity data patterns consistent with arising health issues over a period of time, and the detection of possible emergency conditions passively and unobtrusively. We discuss our efforts using classification techniques leading to the rule-based inference approach, and compare results between the two approaches. The results have shown the viability and validity of knowledge-engineered rules, which outperformed automatically generated rules using random forest supervised learning; the κ correlation coefficient between the classification results of the random forest model and the PDA record was 0.79, with 85 percent sensitivity and 93% specificity, compared to κ=0.84, with 91 percent sensitivity and 100% specificity for the knowledge engineered rule aimed at the detection of main meal preparation. The paper also presents experimental field trial results of the rule-based approach demonstrating the utility of the method and future directions for our research.