Towards Improving Human Activity Recognition Using Artificial Neural Network
Tushar Sharma, P. Kavipriya, Usha Kiruthika · 2021
In medical research and human survey systems, human activity recognition has widespread application. This system leverages the power of smartphone sensors and has proven to be dynamic in various sectors. However, human activity recognition is subject to swift changes in user behavior. This has led to difficulty in prediction of human activities. Previously, Long Short Term Memories (LSTMs) have been used to predict the next instance of activity based upon recent activities. Although LSTM has demonstrated its efficacy in human activity prediction with an accuracy of 92.1% on the HAR UCI Machine Learning Repository, there is a scope for increase in accuracy. This work proposes a robust human activity recognition system by creating a sequential deep neural network. The HAR repository was considered as the dataset to be fed to the proposed model. On inference, the model yields an accuracy of 96.44% on the test set. The results explicate that the model could potentially serve as an aid in the pressing problem of human activity prediction.