Identifying Users Based on Their Activity Pattern Using Machine Learning.

Biometric technology is a very advanced security method that is difficult to duplicate. Recently, fingerprint and face recognition have become popular with the public through their practical use in unlocking smartphones. Research on activity-based biometrics using data from smartphones, smartwatches, and small sensors has been conducted. This study presents a machine learning (ML)-based approach for automatically identifying users based on their daily activity patterns. We used a publicly available human activity dataset, which was collected from eight subjects using on-body three wearable sensors (accelerometer, gyroscope, and magnetometer). First, we extract 21 time and frequency-domain features. Secondly, more efficient features were selected using a minimum redundancy-maximum relevance (mRMR)-based feature selection method. Thirdly, four ML-based approaches, namely random forest, decision tree, 1-dimensional convolutional network, and extra tree (ET) were implemented for user identification. Our experimental results illustrated that the mRMR-based ET classifier produced a higher recognition accuracy rate of 99.5%. Our results indicate that our proposed method can more effectively identify users based on their daily living activities.