Logistic Model Tree for Human Activity Recognition Using Smartphone-Based Inertial Sensors

Human Activity Recognition (HAR) systems using sensor data have widespread use in many real-life applications, making it an important emerging area of research. As inertial sensors are readily available in many handheld devices, HAR systems are generally designed based on the data obtained from them. In this paper, the Logistic Model Trees (LMT) machine learning method for predicting the human motion from smartphone-based inertial sensors is considered. This study aims to demonstrate the capabilities of LMT in obtaining higher prediction rates even with short time segment of data (1 sec), in comparison with longer time segments (2.5 sec) used in the literature. The performance of HAR system designed with LMT is compared with those designed with Random Forest (RF) and Logistic Regression Tree (LR) for a set of dynamic and static activities. The system is trained and tested on two publically available datasets, namely WISDM and UCI HAR. The proposed LMT method outperforms RF and LR by achieving recognition accuracies 90.86% and 94.02% on WISDM and UCI HAR respectively, and achieves between 89.82% - 88.73% overall accuracy during cross-dataset evaluation.