Human Activity and Vehicle Classification using Smart Device Sensors with Machine Learning Assisted Algorithms

Recent research has demonstrated increasing interest in human activity classification (HAC) and vehicle classification (VC). HAR and VC rely on data from smart device sensors like gyroscopes and accelerometers. Traditional HAR and VC procedures often require significant feature engineering and data preprocessing. However, this study proposed machine learning-based algorithms employing decision tree (DT), random forest (RF), and K-nearest neighbor (KNN) classifiers. The experiments used raw sensor data with minimal preprocessing, mainly normalization. Accuracy rates achieved on the HAR dataset for Binary BT Classifier (BDTC), RF Classifier (RFC), and KNN Classifier (KNNC) were 98.20%, 98.21%, and 99.37% respectively. Similarly, for the vehicle dataset, BDTC, RFC, and KNNC were 82.91%, 85.14%, and 99.34%, respectively. Moreover, precision, F1-score, recall, kappa score, and Matthews's correlation coefficient are also evaluated. The results demonstrate the proposed algorithms superior classification performance compared to other published architectures.