Human Activity Recognition System using Smart Phone based Accelerometer and Machine Learning

Human Activity Recognition (HAR) has gained significance importance due to its wide range of applications in security, healthcare, surveillance, virtual reality, control systems and automation. Sensors embedded in modern mobile phones enable unobtrusive detection of Activities of Daily Living (ADL). Various statistical and deep learning techniques for the automated detection of human activity have been presented recently. In this study, we have collected accelerometry data through a mobile phone carried by a user for number of days to classify ADL on the basis of exhibited movement into stationary, light ambulatory, intense ambulatory and abnormal classes. ADL such as walking, sitting and jogging etc. are performed and classified simultaneously by mobile phone application and users for comparative analysis. Collected data is given as an input to the trained model and analyzed by implementing the J48 classifier. Results reveal an accuracy score of around 70% for each activity class and it is noted that the classification was performed with an accuracy of above 80% for stationary activity. It is shown that ADL can be recognized with high accuracy using accelerometry data collected in a constrained environment and a single sensor. J48 classifier also correctly classified activities that have a strong correlation between them such as sitting on a chair and standing in stationary position. This work is significant for utilization in long term health monitoring systems that are capable of ensuring neurological health for masses through HAR and mobile phones embedded with accelerometers.