A comprehensive comparison of machine learning approaches with hyper-parameter tuning for smartphone sensor-based human activity recognition

Abstract

Human Activity Recognition (HAR) using Smartphone sensors like accelerometers and gyroscopes has proven to be one of the most preferred ways to collect data in a simple manner without the need for any additional hardware installation. Smartphones come with these embedded sensors which can be used for monitoring acceleration and angular velocity to help in detecting Activities of Daily Living (ADL). HAR plays an important role in various applications like elderly care, ICU patient care, surveillance systems, etc. Various Machine learning approaches have been applied in the HAR domain to classify human activities using supervised as well as unsupervised learning approaches on various benchmark or compiled datasets. This paper presents a comprehensive comparison of various Machine learning approaches with hyper parameter tuning on a benchmark dataset from the UCI repository. Results prove that the performance of Support Vector Machine (SVM) and Ensemble learning approaches like Random Forest, Gradient Boosting are comparatively better in classifying the activities with hyper parameter tuning using GridSearch as opposed to without applying it for selecting best hyper-parameter values. The performance has been compared with the latest work in the field of Machine Learning in HAR on same dataset and results prove that the proposed approaches outperform most of the existing works in this domain.