Evaluating Residual LSTM approach for predicting missing sensor data for IoMT

Abstract


 
 
 Currently we are living in a digital age due to advancements in the fields like Internet of Things, Artificial Intelligence and Big Data. Especially IoT applications like Smart Home, Smart Watch, Smart Farming, Smart Retail and Smart Parking are seen everywhere around us. These applications are termed Smart due to their ability of self-decision making and monitoring surrounding environment. Data is the most precious assets that requires to make IoT application smart. But multiple times sensor data is found missing or noisy. Due to various reasons like sensor malfunction, sensor maintenance and poor internet communication between devices leads to rise of missing values. In majority cases Missing Completely at Random (MCAR) type missing data are found. Hence, we propose a Residual LSTM model approach for more accurate prediction of missing sensor data on TILES dataset consisting features like Breathing Depth, Breathing Rate and Heart Rate. Compared to approaches like Mean, Median, K-Nearest Neighbours’ and Bidirectional Recurrent Neural Network the Residual Long Short-Term Memory model yields better accuracy which is of huge importance in IoMT application.