A user activity pattern mining system based on human activity recognition and location service

摘要

This poster tries to explore user activity pattern, based on user activity sequences from Human Activity Recognition(HAR) system and locations, which aims to develop novel applications to HAR solutions except existing ones like health care, intelligent homes and so on. Specifically, an Android application is developed to collect inertial sensors data of smart phones. Taking time-frequency domain features and direction features acquired from raw sensor data as input, a Xgboost model is applied to distinguish 8 different activities from user daily life. The experiments show that the HAR system achieves dynamic performance, high efficiency and satisfactory robustness. In the end, several interesting user activity patterns and user properties obtained from user activity sequences are presented.