Anatomical Movement Prediction using Neural Networks
Human Activity Recognition (HAR) using data from wearable body sensors has proved very useful for monitoring the health of disabled or aged people in fast and smart medicare systems. Most of the traditional techniques are centered on finding only one single process that is not efficient for a smart medicare system. Hence, researchers are working for the development of efficient machine learning (ML) techniques where they can use wearable body sensor data to track the activity for smart medicare. In this study, we propose an efficient framework for HAR using wearable sensor data using a RNN model, which uses Long Short-Term Memory (LSTMs). We use LSTM for processing sequential data of sensors by exploiting the capability of their internal states of memory to save the past information and learn from the past information to solve and reduce fluctuations in vanishing gradient. We compared the accuracy with single-layer LSTM with multilayered LSTM, and also with using the conventional recurrent neural networks (RNNs).
Keywords - LSTM, t-SNE, RNN, Wearable Sensor