Classification and Prediction of Low-Density Lipoprotein Cholesterol LDL-C in The Palestinian Patients Using Machine Learning Techniques رسالة ماجستير

Abstract

Cholesterol is one of the major causes of death locally in Palestine and globally in the world, with percentages of 31.5% and 31.4%, respectively. Cholesterol has four main values which are the Total Cholesterol TCH, Triglycerides TG, Low-Density Lipoprotein Cholesterol LDL-C, and High-Density Lipoprotein Cholesterol HDL-C. The main level that is a major factor for Cardiovascular Disease CVD is the LDL-C that is called bad cholesterol, it builds upon the arteries walls narrowing them and slowing the blood flow feeding the heart and brain, causing heart attacks and brain strokes. The science of Artificial Intelligence (AI) and Machine learning (ML) has become a main player that supports in recognizing and diagnosing the LDL-C. The ML depends on the past medical history and heuristic data of the patients diagnosed with Hyperlipidemia. The Hyperlipidemia is the case when the LDL-C exceeding the acceptable and healthy normal ranges threshold of 160 Milligrams per deciliter mg/dL. Classifying and predicting the LDL-C using the ML techniques would guarantee accurate approximation and diagnosis for the disease, avoiding human error, costive laboratory materials, time-consuming waiting for the results. So, the ML has a significant positive effect on the diagnosis and hence the treatment. For this purpose, this thesis utilized ML techniques for classifying and predicting the LDL-C. Additionally, the techniques applied to the HDL-C classification and prediction. The utilized techniques are the Artificial Neural Networks ANNs, Recurrent Neural Networks RNN, Radial Basis Function Neural Networks RBFNN, Fuzzy Logic, Support Vector Machines SVM, Decision Tress DT, Logistic Regression LR, and a hybrid model of combining the ANNs with Fuzzy Logic for optimizing the results accuracy and reducing the error. These methods require a dataset for training and testing the used techniques. Since the VI study targets the Palestinian community of Cholesterol patients, so cooperation with the Palestinian Ministry of Health MoH had taken a place, they provided the needed medical history data and risk factors related to the Palestinian patients of Cholesterol. Another additional supportive international dataset is utilized, which is collected by the Korean National Health and Nutritional Examination Survey KNHANES, it's used to generalize the results and compare with the other efforts. The used and produced methods outperformed the other efforts done on the same idea with a significant difference, the older efforts done in the research titled "Lipid profile prediction based on artificial neural networks", reached 80% accuracy as they relied on a fewer number of fields limited to the lipid profile values only, where in this thesis it reached to the accuracy of 97.10% in the international dataset and 95.55% in the national dataset. Though this thesis recommends including more affecting risk factors and fields, especially the ones related to the noninvasive works, as it will lower the cost, efforts, time, and increase the accuracy