Human activity recognition utilizing SVM algorithm with gridsearch
Wahyu Andhyka Kusuma, Agus Eko Minarno, Nia Dwi Nurul Safitri · AIP conference proceedings · 2022
Human Activity Recognition (HAR) has become one of the most debatable research topics due to the availability of sensors and accelerometers, the low cost and less power consumption, the live data streaming, and the advances in computer vision, machine learning, as well as in artificial intelligence. This study proposes a HAR system utilizing Support Vector Machine (SVM) algorithms to navigate hyperplane which could separate the two datasets from two different classes. Hyperplane is defined as a function separating between classes. Regardless of its capability for data classification, such an application enables this research to be distinct from previous research by adding the application of Tuning Hyperparameter to SVM. Further, the Dataset which initially contains the combined data, is separated into a static activity dataset and a dynamic activity dataset. In this study, the accuracy of applying SVM+Hyperparameter has raised up to 96.26%. Hence, this selection becomes the novelty of this research, which could improve the accuracy from previous research without applying Hyperparameter.