Novel Machine Learning Algorithms For Predicting Risks Of Chronic Diseases

Abstract

Personalized prognostic models have been used for predicting the risk of chronic newlinediseases, with the adoption of electronic health record (EHR) systems. Most of the newlineworks in this domain has been on binary classification to predict the onset of one newlinechronic disease with little attention given to comorbid conditions and mortality newlineassociated with it. Precision of the prediction model in medical informatics can be newlineimproved by leveraging the advanced concepts of machine learning and deep learning newlinealgorithms. To a large extent, this is not easy because of the multiple facets of patient newlineinformation being encoded in the data. Hence, it is important to derive a correct newlinerepresentation of the EHR encodings to enable its usage in analytical tasks, efficiently newlineextract the knowledge from the data and then use that knowledge in machine learning newlineand deep learning model to get more efficient and accurate prediction of risk of the newlinechronic disease and clinical outcomes. To address the identified gaps on this context, newlinethree new approaches are introduced. First, a parameter agnostic framework for newlinepredicting disease onset using deep learning methods is designed. Then a newlineHeterogeneous Recurrent Convolution Neural Network (HRCC) is developed. Finally, newlinea Multi-Level Spatial Coherence Optimization Approach (MLSCO) is modeled and newlineimplemented for better results. newlineIn this research work, focus is on the prediction on chronic disease risks. EHR is used newlinelargely as data source along with other supporting data sets for knowledge extraction newlineand value mappings. Multiple deep learning techniques are implemented in this newlineapproach. Through experiments, it was proven that individual feature labeling is low in newlineaccuracy and performance. Considering several feature variables and rules will improve newlinethe accuracy and will result in better efficiency for predicting the risk. Evolution of newlinethese algorithms with multiple experiments results have showed the highest accuracy newlineof 97.44%. By developing novel algorithm for prediction, a generic framework is newlinedesigned by combining multiple algorithms, which is used for finding disease newlineprogression and risk prediction. Therefore, this proposed method and combination of newlinerules and features performs better in risk prediction of chronic disease. newline