Evaluation of Feature Extraction and Recognition for Human Activity using Smartphone based Accelerometer data

Fall is one of the major life threatening problem faced by elder, because falls are dangerous and may result in death sometimes due to lack of help after fall. In the majority of fall events, automatic Ambient Assistive Living provides essential support to avoid major consequences. This may reduce the response time and significantly improves the prognosis of fall detection. This paper presents a fall detection system using Statistical and Higher order statistics features evaluated through Recursive Feature Elimination and classified with J48 and k-Nearest Neighbor algorithms for the tri-axial accelerometer data. The accelerometer data has been used from the popular smartphone based fall detection dataset MobiAct. The proposed system performs better with minimal number of features and depends only on accelerometer data rather state-of-the-art technique depends on more data such as Accelerometer, gyroscope, orientation with more features for better accuracy.