An Approach to Handle Heterogeneous Healthcare IoT Data Using Deep Convolutional Neural Network

Abstract

Internet of Things (IoT) is going to be the next big technological revolution of mankind by connecting everything on the earth via the Internet. Mobile healthcare or remote healthcare is an important application of IoT, which provides a new platform to people for getting benefit regarding healthcare-related problems. In these applications, various sensing devices connect with the patient’s body and generate an enormous amount of heterogeneous data over time. Due to the variety of data, extraction of knowledge from these data is not straight forward just like the conventional data mining process. In this research, we have proposed a Deep Convolution Neural Network (DCNN)-based classification method for performing data mining over heterogeneous data by taking an unstructured sensor dataset from the arrhythmia database of physionet. In the proposed method, the CNN feature extraction layer converts the ECG signals into numeric form by calculating its features without any human intervention. Similarly, categorical data are converted based on their respective categories. Finally, all the converted data together were added to the CNN classification part and arrhythmia disease is predicted with an accuracy of more than 98%. The simulation result shows that our proposed CNN-based architecture outperforms other handcrafted feature extraction techniques in terms of accuracy, sensitivity, and specificity.