Human Activity Classification Using Machine Learning Techniques with Feature Selection

Abstract

Recognition of human activities is a field of rising interest in the present transformation. Recognizing the activities allows the applications in constructing the activity profiles for every subject that could be used efficiently for the health and safety applications. Nowadays smartphones are upgraded with remarkable motion sensors to revitalize the entryways in machine learning. According to human activity recognition, datasets for the corresponding work have been increased that paves the way for certainties in different research areas. The human activity recognition dataset includes accelerometer sensor data with respect to six different human activities for example sitting, standing, lying, walk, upstairs walk, downstairs walk. The paper presents the human activity recognition process by analyzing the filter-based feature selection methods in the machine learning model. For activity classification, various classifiers are used as Support Vector Machine (SVM), K Nearest Neighbor (KNN), Random Forest and Decision Tree classifiers. The performance of the model is validated using Train/test split approach. Therefore, for activity recognition, we proposed a model giving better results with a random forest classifier and correlation-based feature selection with best-first search.KeywordsAccelerometerCorrelation-based feature selectionHuman activity recognitionKNNSVMRandom forest