Random forest application in physical activity categorization for users of mobile devices

Abstract

The purpose of this paper is to introduce the classification problem on the base of physical activity categorization in the terms of differentiating several types of activities based on data gathered from mobile devices sensors. The solution in the form of expert system with machine learning interface engine is proposed. Random forest theory with fundamentals and key features is described including single decision tree. Classification and Regression Trees algorithm has been introduced. In the practice part of this paper all the theoretical knowledge has been used in the script created in Python programming language featuring random forest classifier. The external dataset and data collected during research containing sensor data from walking and running are tested in this script in various way including mixed frequency and sensor type. The goal of the categorization was to predict the activity type based on single sensor sample. Script performance and output is described including single decision tree with all nodes.