Future Human Activity Prediction Using Wavelet And Lstm
Mohammad Khalaf Rahim Al-juaifari, Alih. Ali Athari · The Journal of The University of Duhok · 2023
Future Human Activity Prediction holds significant importance as it enables early detection and monitoring of various aspects, such as elderly care, early fall detection systems, smart-home applications, and E-health monitoring.A pioneering approach has been developed to achieve this, incorporating the Wavelet transform preprocessing technique for dimensional reduction through signal decomposition.This is followed by the implementation of a deep learning model supported by time series data, enabling real-time monitoring of physical activity.A novel method has been proposed based on wearable sensor data sources, employing LSTM and time series models, and applied with MHEALTH Dataset.This dataset comprises 12 complex activities and sensor-based devices, ensuring the privacy of patients or participants in real-life scenarios.The results demonstrate that the predicted activity of five steps with an accuracy level for the next day's activity achieved an accuracy of 98%, surpassing the accuracy and complexity compared with state-of-the-art methods.