Genetic Algorithm Optimized BiLSTM Networks Model for Human Activity Recognition using Timeseries Sensor Data
S. Selvabharathi, K. S. Dhanalakshmi, P. Prabhu · 2023
In recent years, Human Activity Recognition (HAR) is a method used to classify human physical activities in a given period of timestamp based on accelerometer and gyroscope sensors collected data. It plays vital role in various applications like assistive living to monitor individuals’ behavior, training & monitoring employees and health & fitness monitoring with the use of various types of sensors present in mobile phones and wearable devices. This system is used to detect the abnormal behavior in assistive technology. Deep learning provides an opportunity to solve many issues in various sensor-based applications. In this work a hybrid method which combines the benefit of Genetic Algorithm (GA) and bidirectional LSTM is proposed. This machine learning (ML) model results in better accuracy up to 97.3 and 98.8% for UCI smartphone and WISDM public datasets respectively and robustness in recognizing real time activities.