Development of Activity Recognition Model using LSTM-RNN Deep Learning Algorithm

Abstract

This study analyses numerous human activities and also classifies the activities based on their trait of motion using wearable sensors data. As a part of the Human Activity Recognition Framework's evelopment, the LSTM-RNN algorithm was implemented. We have considered ten types of motions for recognition and based on the duration of motions have classified those motions into repetitive and non-repetitive motions. The dataset utilized to evaluate the model's performance was recordings from Opportunity. The best trained model achieved an overall accuracy of 94% and The findings of the study stated that the LSTM-RNN model achieved greater accuracy of 91% pertaining to motions that are not repeating that means motions that are performed for short periods of time in comparison to the motions having long dependencies which achieved accuracy of 80%. The determination of performance has been done in terms of score of accuracy, score of precision and f1 score. In addition to this, a disparity analysis of the presented model with another devised model has also been done.

Keywords

• Long Short Term Memory
• Recurrent Neural Network
• Human Activity Recognition
• Sensor
• Accuracy