Indoor And Outdoor Mobility Classification System Based On Android Phone And Machine Learning

With the improvement of people's living standard,people are increasingly caring about their health.For instance,always driving car will increase the emissions of CO2 that will pollute the environment;often sitting down for a long time will incur lumbar disc protrusion.In order to create a green life,reduce environment pollution and make people lead a healthy life,create a system that can detect and classify people's mobility and help people live a green life through reward and punishment rules is of great significance.This paper realized a reward and punishment rules based android application that uses GPS and its in-built sensors accelerometer and gyroscope to collect users' data and use machine learning algorithms to classify outdoors and indoors activities.Outdoors activities include stationary,walking,running,taking a bus,taking a tube,and riding a bike.And indoors activities include stationary,walking,sit down,climb up stairs,climb down stairs,take a lift up,take a lift down,fall down and Parkinson's disease.The system first uses GPS,accelerometer and gyroscope to collect users' data.In terms of feature selection,we investigate how to better choose features that can make better classification.This paper proposed,by using accelerometer,choose some time domain features such as peak and trough features,statistical features,steps and Wi-Fi.Through gyroscope,the highest and lowest skew of the angular velocity are also chosen.Then,supervised machine learning classifications algorithms are used which include decision tree,random forest,SVM,neural network,na?ve bayes,KNN,and logistic regression.Comparisons are made between using raw data and its median to do classification.Experiments show that median of the raw data is not as good as the raw data in terms of classification input data.This paper also investigates how different combination of GPS,accelerometer and gyroscope and the change of parameters of machine learning algorithms can improve classification accuracy.By doing experiments,the algorithms are compared and the better classification algorithms are selected,which are random forest,decision tree and neural networks.Experiments indicate that the use of GPS,accelerometer and gyroscope can obtain best classification results.In outdoors activity classification,random forest achieved 88.57% accuracy;in indoors activity classifications,decision tree obtain 97.54% accuracy.Number of users is increased from 5 to 50 and hypothesis test is used to validate the representativeness of users' data and feasibility of classification algorithmswhen the sample size is small.The system can improve the accuracy of detecting Parkinson disease and fall down.And make reward and punishment rules in android phone and also using rule based engine Drools to make rules on server side to better help people lead a healthy life.This system is of great importance to people's health.It uses smart phone and machine learning algorithms to make mobility classification with high accuracy.And through reward and punishment rules,it can lead people a healthy and green life and is beneficial to sustainable development.