Classifying Users’ Exercise Profiles Based On Mobile Platform’s Exercise Data

Nowadays,more and more researches show that the exercise is good for health.Under the guidance of national policy,consciousness of the public in participating in exercises is greatly enhanced.At the same time,the arrival of the era of mobile Internet has greatly promoted the development of online sports platforms,and a large number of software and hardware used to record the user’s activity data have emerged.Step data of exercise(SDE)is a kind of time series triggered by human movements and recorded by mobile devices.The collection of SDE is becoming more and more popular because it provides users with information about the level of activity,and it can provide stimulation for users’ active lifestyle.When SDE becomes more widely available,it also provides a new way for health professionals to interact with patients.However,the problem of obtaining user’s exercise habits from the user’s SDE data recorded by these software and hardware and understand the user’s exercise habits from the user’s SDE data recorded by these software and hardware has not been thoroughly studied and understood.The purpose of this paper is to study this kind of step data generated from user’s exercise,and design a new classification algorithm to identify user’s exercise modes.The few existing methods of classification rely on manually labeled training models and often need template,or require multi-dimensional sensor data which is expensive to obtain.However,SDE data sets are highly personalized.And due to technical limitations,such data is not completely accurate.For those reason,the cost and complexity of the existing analysis methods are typically very high.In this thesis,we study the classification of SDE.Instead of using the raw steps data as the input for classification,we try to interpret the data from an exercise behavior point of view,and design a noise reduction filter to reduce the interference of noise that may be generated by daily routines other than exercises.At the same time,we extract the prior knowledge of the user’s exercise preference based on the user’s personalized characteristics,and then divide the initial SDE time series into time windows.The resulting classification algorithm has lower complexity and labor cost compared to existing ones.Furthermore,this paper evaluates the proposed classification algorithm in many aspects,including accuracy,robustness,aggregation of classification results,etc.,and compares it with other algorithms in these aspects.The test results on two data sets prove that the proposed algorithm has significant advantages in the classification of SDE time series.Our method provides easy to understand and reliable feature description for users’ exercise habits.While maintaining low computational complexity,it significantly improves the accuracy of classification.At the same time,this paper also inspected the exercise behavior of the user group,which may help us to understand individuals.The classification algorithm proposed in our research can be applied in various platforms where SDE is collected.It provides us with a robust tool to understand the user in terms of daily activities and lifestyle.