A Novel Procedure for Recognising Human Activity Using Machine Learning

In the past few years it is seen that there has been a boom in the usage of smart devices consisting of sophisticated sensors such as gyroscope or gyro and accelerometer. Human Activity Recognition (HAR) is one of these sensors' applications. Given the dependence on general Machine Learning (ML) techniques, Deep learning techniques are suggested by current research to be suitable for feature extraction from unprocessed sensor data. The study focuses on Human Activity Recognition (HAR) by means of data gathered from smartphone accelerometer sensors and employs Long-Short Term Memory (LSTM) networks in the time-series domain. Furthermore, a hybrid network consisting of a 2-layer RNN-LSTM is proposed to enhance the accuracy of recognition performance. For the input and output activation function in the neural network ReLU and Softmax function respectively are used.