An Efficient Kernel KNN classifier for Activity Recognition on Smartphone

The real-life mobile sensing applications use mobile sensors integrated in smartphones to predict user's physical activities, and to detect an anomaly. The precision of the human activity recognition (HAR) system depends on extracted features and robustness of the training model. This study mainly proposed a new scheme SV-KNN based on kernel K-Nearest Neighbors using a compact training data based on the support vectors (SV) to identify the ongoing activity. To reduce the sensory features as the inputs for classifier, we used the Principal Component Analysis (PCA). Comparison of our system with existing classifiers shows the efficiency of SV-KNN approach in terms of accuracy and F-score.