Prediction of Chronic Kidney Disease Depending on Patient Treatment History Using Machine Learning Techniques رسالة ماجستير

Abstract

Chronic diseases are among the most consuming diseases of health care budgets and are characterized by treatment difficulty. Chronic kidney disease (CKD) is a global health problem with a high morbidity and mortality rate. Since there are no visible symptoms during the early stages of CKD, CKD is one of the diseases that have no efficient treatment till now, and the cost of treatment is high. Prevention and prediction are the two ways that can help in improving CKD patients' health and reducing the treatment costs. Early detection of CKD allows patients to receive timely treatment, which improves the progression of this disease treatment. Machine Learning (ML) models can effectively help health professionals to achieve this goal due to their fast and accurate recognition performance of the CKD. In this study, we proposed a technique based on Machine Learning methodology for the diagnosis of CKD. Different ML techniques were used to classify and predict CKD using Patients' treatment history data collected from the Palestinian Ministry of Health (PMOH) repository. First, a preprocessing step was used for cleaning, transformation, and feature selection for the collected dataset. In the next step, we applied Eight ML algorithms, namely: Decision Forest (DF), Decision Jungle (DJ), Support Vector Machine (SVM), Locally Deep Support Vector Machine (DSVM), Logistic Regression (LR), Boosted Decision Tree (BDT), Bayes Point Machine (BM), Neural Networks (NNs). These ML models' performances will be compared to decide the best classifier model in predicting CKD for the given dataset. vi Therefore, the research method depends on a list of steps, starting by collecting the data, data preprocessing, determining the prediction and classification factors, training and testing data will be applied, and implementing different ML models using the training and testing dataset to classify and predict CKD. The dataset worked entirely in the cloud on the Microsoft Azure platform. The results of the models used showed the superiority of the model Boosted Decision Tree. It outperformed other models with the following results: Accuracy= 0.945, Precision= 0.95, Recall= 0.95, F1Score= 0.95, and AUC= 0.972. The experimental results indicate that using the ML techniques to predict and classify the CKD supports the early diagnosis before the disease grows to advanced stages that are difficult to treat. Many patients can avoid kidney dialysis or search for any donor to perform a kidney transplant.