Missing data handling using machine learning for human activity recognition on mobile device

Abstract

Human activity recognition is important technology in mobile computing era because it can be applied to many real-life, human-centric problems such as eldercare and healthcare. Successful research has so far focused on recognizing simple human activities. Currently, the smartphone is equipped with various sensors such as an accelerometer, gyroscope, digital compass, microphone, GPS and camera. The sensors have been used in various areas such as human gesture and activity recognition which is opening a new area of research and significantly impact in daily life. Activity recognition between the personal computer and smartphone is different. A mobile device has limited computational and memory capacity which has a chance that some data are missing when limitation of the mobile device is happening. In this research, some algorithms are tested to perform their ability to handling missing data, they are Bayesian Network, Multilayer Perceptron (MLP), C4.5 and k-Nearest Neighbour (k-NN). Missing data are implemented with increment scaling from 5%-40%. Optimal result based on accuracy mean is obtained by kNN with 89,4752%. Based on class, Bayesian Network obtained mean 992 recognized on Sitting class and kNN obtained mean 1010 recognized on Walking class. Multilayer Perceptron is obtained endurance point with decreasing about 9.9109% from normal experiment without missing data.