Activity Detection and Analysis Using Smartphone Sensors

The smartphones have multiple built-in sensors – each having a specific function which helps the device perform efficiently. These sensors can potentially be used to detect events or gather data about human activities. The use of machine learning algorithms on smartphone sensor data could have numerous applications in finance, healthcare, entertainment, etc. In this research, we develop mechanisms to gather sensor data and study how the data can be used to differentiate patterns in user physical motions. In this research, multiple behaviors were distinguished using sensor data. For our experiments, we focus on three classes of human activity which are walking, standing, and running. Included in our data is the sensor input while the device is sitting in a stationary position. We developed a smartphone application for the Android platform to gather data from a mobile device. To detect the device motion, our experiments will use the smartphone accelerometer and gyroscope. From the data collected through an application, we create a feature set which consists of linear acceleration, normal acceleration and angular acceleration of the device. To differentiate each activity in our dataset, we used a naive Bayes algorithm and k-means clustering. Our model showed an 85% accuracy in activity classification using a data set of five hundred sensor data entries. In conclusion, our results show that onboard sensors of smartphones can contribute in human activity recognition.