Hypermeters Optimization in Recurrent Neural Networks-LSTM Approach for Human Activity Recognition

Abstract

Human activity recognition (HAR) has become more popular with the increase of applications involving human-computer interaction. The problem of recognizing and classifying people's daily life activities is a very important and challenging issue in the field of artificial intelligence. In recent years, deep learning based techniques have achieved high accuracy for HAR. However, these methods require obtaining a optimal number parameters. In this study, we investigated Long Short-Term Memory (LSTM) for HAR from videos. In addition, hyperparameters of the model were obtained thourh a thorough search to optimize ther performance of the model. Specifically, we obtained optimal values for number of layers, batch size, epochs to obtained best accuracy for the model. Experiments were performed to evaluate the performance of the proposed model on WISDM dataset. The proposed model produced an overall 97.18% overall accuracy indicates that LSTM is an effective technique for HAR.