An Efficient IOMT Based Disease Prediction Model

Abstract

Chronic disease is one of the most common causes of death worldwide. Lack of awareness and inability to access medical help lead to complications in such patients. In IoMT (Internet of Medical Things), different sensors collect patients’ vitals that can be utilized by expert systems to predict health concerns. However, data from these sensors is prone to inconsistency, and imprecision thus affecting the model in predicting accurate results. Existing technology utilized pre-processing techniques and different feature selections for cleaning that data followed by different machine learning (ML) models. Literature reveals that optimization techniques are costly in terms of time and have lesser accuracy. Thus we have proposed an ensemblebased, technique with relevant feature extraction and pre-processing methods. We utilize the Heart disease dataset and kidney disease dataset to evaluate a proposed technique stacking and voting classifier that generates the highest accuracy other than individual models. Also have compared the results of the ensemble technique with the optimization algorithm results. Stacking evaluates the 95% accuracy, 94% precision, 97% re-call, and f1-score 95% on the kidney disease data set. Also, the hard classifier evaluates the 84% accuracy, 88% precision, 87% re-call, and f1-score 84% on the heart disease data set. The ensemble classifier achieved higher accuracy than the optimization algorithm.