Multi group based daily living activity recognition (DLAR) using advanced machine learning algorithm

Abstract

Human activity recognition (HAR) has realized more interest in several research communities given that understanding user activities and behavior help to deliver proactive and personalized services. Different types of noise in wearable sensors data frequently hamper the human activity recognition classification process. Categorically, in this proposed work designs the three-level hierarchical classification structure, i.e., instance based, group and sub-group based and subject based to detect the daily activity of human body motion among activity groups. In correlation with other famous classifiers, for such as Random Forest Tree, J48, Decision Table, Multilayer Perceptron, NaïveBayes, oneR and REPTree (Reduced Error Pruning Tree), etc., thorough experiments on the mHealth dataset (Shimmer2 mHealth Data) demonstrate that group based classification achieves the best classification results, reaching RFT 99.97%. We trained classier in order to estimate accuracy classification based on (gender, age, height, and weight). We applied validation methods to the process, 10-fold cross-validation. For all three classification structure, we achieve high accuracy values for all three classification task.