Application of a Cluster-Based Classifier Ensemble to Activity Recognition in Smart Homes

An increasingly popular technique of monitoring activities within a smart environment involves the use of sensor technologies. With such an approach complex constructs of data are generated which subsequently require the use of activity recognition techniques to infer the underlying activity. The assignment of sensor data to one from a possible set of predefined activities can essentially be considered as a classification task. In this study, we propose the application of a cluster-based classifier ensemble method to the activity recognition problem, as an alternative to single classification models. Experimental evalua-tion has been conducted on publicly available sensor data collected over a period of 26 days from a single person apartment. Two types of sensor data representation have been considered, namely numeric and binary. The results show that the ensemble method can predict a numeric and binary representative activity with accuracies of 94.2% and 97.5%, respectively. These results outper-formed a range of single classifiers.