Human Behavior Analysis And Prediction Based On Sensor Data In Smartphone

The development of modern industry and technology has brought not only many conveniences to people's lives,but also great changes in people's living habits.Studies have shown that many chronic diseases are associated with long-term unhealthy habits.In order to prevent these chronic diseases,we must first understand and analyze people's daily behavior,and then provide interventions to change unhealthy behaviors.In this article,we focus on the collection,analysis,and predicting of people's behavior data.We describe the key technologies for data collection and the data collection software we design.For related work on user behavior data processing technology,a detailed introduction is provided.Based on the data we collect and other public available datasets,we use machine learning techniques to identify user actions such as walking,going up and down the elevator,running,etc.The algorithms we use are KNN and SVM.KNN is an efficient algorithm,while SVM is more sophisticated.One of the research questions in this paper is whether KNN and SVM can effectively identify user behavior under our dataset.Based on users' behavior recognition and user's geographical location change,we can generate the spatio-temporal data of users' behavior.We compared the performance of two kinds of sequence prediction techniques(first-order Markov chains and ActiveLeZi)on user behavior and user geographic location prediction.The first-order Markov chain is the simplest Markov model,which assumes that the current state is related only to the state of the previous moment.Active-LeZi is a more sophisticated Markov model.According to the patterns in the data,the current state may be related to multiple continuous states before.The second research question is to compare these two models in behavior and location prediction.Our experimental results show two findings.Firstly,KNN and SVM perform different on user behavior recognition for the two datasets we used and one can beat the other on each dataset,respectively.Secondly,first-order Markov chain is superior to Active-LeZi in predicting motion behavior,while in user location prediction,both methods have chance to win.In general,the used machine learning techniques have a good effect on the defined application scenarios.Finally,we discuss the main contributions and limitations of the paper,analyze the application prospects of user behavior intervention,and summarize the whole paper.