MachinelLearning to classify and recommend physical activity based on wearable technology

Abstract

Different wearable technology has been used to measure the physical activities of people with movement disorder through activity classification as well as recommending suitable physical activity level. A physical activity is important for maintaining healthy lifestyle. Although people are aware of health-related guidelines, physical activity among individual is declining. Wearable devices are becoming popular within individuals for observing own activity. By observing those data, it helps us to follow various guidelines of our activity. With the possibility of wearable computing there are various recommendation a person should perform based on their physical health. Some studies have shown the effectiveness of wearable devices for increasing physical activity level and improving health based on their health condition. Therefore, identifying the physical activity and recommending activities is important for healthy and prosperous life. Demand for wearable devices increased and the research area in the field of wearable computing also increasing. The data generated by sensors gives a distinctive scope to understand the user behaviour and pattern. It has a potential to improve our health level. In this study, we identify and classify the user behaviour on different activity with the help of machine learning model and provide recommendations for users regarding the physical activity. We use four different supervised machine learning algorithms to obtain best predictive accuracy. We use correlation analysis and Principal component analysis (PCA) for maximizing variance. K Nearest Neighbor, Random Forest, Decision Tree and Logistic Regression are used for classification. We use cross validation technique to perform the evaluation of models. We use unsupervised machine learning algorithm to provide recommend new activities. The Clustering Algorithm k-Means is used for recommendation