Multimodal Home Monitoring of Elderly People--First Results from the LASS Study

Monitoring elderly or disabled people in smart home environments is a major area of research because it allows for controlling chronic diseases and promises cost reduction. Context recognition and in particular activity recognition is of key importance as it facilitates the interpretation of data from medical monitoring devices. In our study with five elderly or disabled people we used data from multi-sensor wearable devices to generate intra- and interindividual machine-learned classifier models to determine activity patterns. Furthermore we computed the relative relevance of each parameter measured, and assessed the acceptance of computerized questionnaires in computer- illiterate people. The mean classification accuracy was 91.4% for the intraindividual classifiers and 53.7% for the interindividual ones. The most relevant parameters for activity classifications were those derived from accelerometric data, the least relevant one was galvanic skin response. Both the sensor device and the computerized questionnaires were well-received by the study participants. Individually-trained machine-learned classifiers used on data from a wearable device are an adequate means to determine context in elderly or disabled people.