Deep learning framework for analysis of health factors in Internet-of-Medical Things

Abstract

The introduction of IoT technologies, such as those used in remote health monitoring applications, has revolutionized conventional medical care. Furthermore, the approach utilized to obtain insights from the scrutiny of lifestyle elements and activities is crucial to the success of tailored healthcare and disease prevention services. Intelligent data retrieval and classification algorithms allow for the investigation of disease and the prediction of aberrant health states. The Convolutional Neural Network (CNN) strategy is utilized to forecast such anomaly because it can successfully recognize the knowledge significant to disease anticipation from amorphous medical heath records. Conversely, if a fully coupled network-topology is used, CNN guzzles a huge memory. Furthermore, the complexity analysis of the model may rise as the number of layers grows. Therefore, we present a CNN target recognition and anticipation strategy based on the Pearson Correlation Coefficient (PCC) and standard pattern activities to address these shortcomings of the CNN model. It is built in this framework and used for classification purposes. In the initial hidden layer, the most crucial health-related factors are chosen, and in the next, a correlation-coefficient examination is performed to categorize the health factors into positively and negatively correlated groups. Mining the occurrence of regular patterns among the categorized health parameters also reveals the behavior of regular patterns. The model output is broken down into obesity, hypertension, and diabetes-related factors with known correlations. To lessen the impact of the CNN-typical knowledge discovery paradigm, we use two separate datasets. The experimental results reveal that the proposed model outperforms three other machine learning techniques while requiring less computational effort.