Predicting the risk of hypertension with machine learning

Hypertension is the leading risk factor among preventable causes of death worldwide. Hypertension usually occurs asymptomatic. In people with hypertension, exposure to constantly elevated blood pressure often causes undesirable conditions such as arteriosclerosis, myocardial infarction and stroke. Hypertension, a major global health risk, affects 1.13 billion people worldwide. However, almost half of individuals do not know that they have hypertension. Therefore, it is important to improve the diagnosis and monitoring of hypertension to reduce the likelihood of cardiovascular diseases (CVD) and to keep blood pressure values at an optimum level. Blood values, physical examination findings, urine tests, and home blood pressure monitoring are generally used to diagnose this disease. Apart from these diagnostic methods, computer-aided programs are used for the diagnosis of this disease, as in many areas. The aim of this thesis is to diagnose hypertension at an early stage, which is a disease that is asymptomatic until the later stages, usually presents with complications, and complications can be prevented if detected early. By using machine learning algorithms, hypertension can be diagnosed at an early stage, and by directing the individual to the correct and early treatment, thus preventing late diagnosis and preventing the individual from being diagnosed. It is to protect against the risks of other diseases that come with it and to help them live a comfortable and comfortable life. The data set in this study was prepared from the data obtained from patients who applied to Isparta Süleyman Demirel University Family Medicine Polyclinic between 01.01.2019 and 01.01.2023. The data set consists of a total of 399 people. This data set consists of 34 parameters. One of them is a class variable. In the study, four different machine learning algorithms were used to detect HT at an early stage. The performances of these algorithms were evaluated with the WEKA package program. The data was analyzed by dividing it into 66% training set and 34% test set. As a result of the analysis, the SVM (Support Vector Machines) algorithm gave the best performance with an accuracy rate of 80.147%. TP ratio: 0.801, FP Ratio: 0.202 and ROC Area: 0.800. The findings of all algorithms applied in the study are shown in detail in the tables included in the study.