Feature Selection and Hyperparameter Optimization of SVM for Human Activity Recognition

Activity recognition has received a lot of attention from research scholars in the past few years. There has been a huge demand for activity recognition because of its ability to ease human-machine interaction, help in care for the elderly, and monitor the habitat requirements of the wildlife. In this paper, a Support Vector Machine (SVM) classifier to recognize the human activities has been built. Data was collected from the database provided by the University of Southern California (USC) for human activity recognition. Six features were computed to obtain the feature set. Different feature subsets were then evaluated based on the precision and recall scores. Using grid search algorithm, the best subset of hyperparameters (SVM kernel, regularization parameter(C) and γ) for the SVM classifier which gives the highest precision and recall score was selected from the parameter space. The best set of features and the SVM hyperparameters for obtaining best results in activity recognition are proposed in this work.