Soft margin SVM modeling for handling imbalanced human activity datasets in multiple homes

Activity recognition datasets are generally imbalanced, meaning certain activities occur more frequently than others. Not incorporating this class imbalance results in an evaluation that may lead to disastrous consequences for elderly persons. In this work, we evaluate various types of resampling methods: at algorithmic level using CS-SVM and at data level using SMOTE-CSVM and OS-CSVM combined with the discriminative classifier named Soft-Margin Support Vector Machines (CSVM) in order to handle imbalanced data problem. We conduct several experiments using three real world activity recognition datasets and show that the SMOTE-CSVM and OS-CSVM are able to surpass CRF, CSVM and CS-SVM. OS-CSVM is slightly better than SMOTE-CSVM for classifying the activities using binary and ubiquitous sensors.