Enhanced Human Activity Recognition (HAR) with IMU Sensors in Smartphones: Insights from Machine Learning Models

Human Activity Recognition (HAR) is the main software component in healthcare, sports, and interactive mobile applications. The accuracy of HAR component is strongly tied with the sensor used for the motion detection and it dictates the overall performance of the application. This paper investigates the use of Inertial Measurement Unit (IMU) sensors embedded in smartphones to investigate the HAR accuracy through machine learning approach. The accelerometer and gyroscope outputs are utilized to classify six human activities: going downstairs, going upstairs, sitting, standing, walking, and running, using Recurrent Neural Network (RNN), Random Forest (RF), and Deep Learning (DL) algorithms. A time series dataset comprising XYZ-axis measurements from accelerometer and gyroscope sensors across four types of smartphones, involving 30 participants is used to train the machine learning (ML) models. To enrich the dataset, sensor filtering, and fusion techniques are employed to evaluate different scenarios. The findings of the study provide significant insights into the capabilities of smartphone-embedded IMUs for HAR in mobile applications.