<![CDATA[Understanding behavioral patterns and forecasting the bodily motions of persons heavily relies on detecting human activities. This has profound ramifications in several domains, including healthcare, sports, and security. This study sought to identify and classify 18 human actions recorded by 90 people using smartphone sensors using the KU-HAR dataset. The primary aim of this study is to examine statistical features such as (mean, mod, entropy, max, median …etc.) derived from time-domain sensory data collected using accelerometers and gyroscopes. Activity detection utilizes many machine learning methods such as Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Logistic Regression (LG), Naïve Bayes (NB), and AdaBoost. The RF model achieves the highest overall accuracy of 99%. While the DT model gets 95%, SVM receives 93%, and the KNN gets 82%. At the same time, the other model didn’t get good results. The research is evaluated using accuracy, recall, precision, and f1-scor. The research contribution is to extract the statistical feature from the raw file of the sensor to create a new dataset. This research recommends employing statistical features in time series. Future research is recommended to solve misclassification in certain activities, which could be achieved using feature selection to reduce the number of features.]]>
Copyrights © 2024
