Context Analysis and Estimation of Mobile Users Considering the Time Series of Data
Hiromi SHIMIZU, Mutsumi Suganuma, Wataru Kameyama · 2020
Recently, the demand for understanding mobile user's activities has been increasing in the fields such as customer behavior analysis. Our previous study shows that mobile user's context can be estimated to some extent using various sensor data of mobile phone and user's bio-signals by applying machine learning methods. In this paper, we propose to analyze and estimate the context by considering the time series of data to improve the accuracy. For the analysis and the estimation, using the data collected from two subjects, convolutional neural network (CNN) and various machine learning methods to classify the data into pre-defined eight and seven contexts are applied and compared. The results show that CNN with 256 window width achieve the highest macro F1-score of 97.6% and 97.1% for each subject, respectively. It suggests that the context analysis and estimation using sensor data and bio-signal can be done much more accurately by considering the time series of data.