Prediction of Human Physical Activities With Consumer Smartphone Sensor Data

Abstract

Sensor-dependent recognition of human activity (HAR) is an exciting and challenging research topic. It is commonly used in many areas, such as sports, health, criminal investigations, etc. Billions of people now have smartphones with sensors such as accelerometers and Gyroscopes that can easily record data related to a person’s activity. Those data can be helpful to build a monitoring system. Classical machine learning algorithms use human-engineered feature data. However, it requires domain expertise and signal processing techniques to get the best feature from the time series data recorded from sensors. It takes a tremendous amount of time and effort to get the best features for human activity recognition. But nowadays, the recent advantages of deep learning make it easier for everyone to detect human activity faster and more accurately. Deep learning models can be used to construct a suitable monitoring system for human activity prediction. This paper proposed a Deep LSTM (long short term memory) model for human activity recognition. The performance of the proposed model is evaluated in terms of accuracy and F1 score. The proposed approach requires lesser number of parameters for real time activity detection.