Abstract:

Heart diseases are ranked first among all diseases in terms of mortality. Although there is no precise treatment for the disease, the correct diagnosis affects the patient’s survival time and quality of life in a positive way. Many clinical methods have been used for the diagnosis of heart disease. In recent years, machine learning algorithms are also used to diagnose the disease. In this study we have used the KNN classifier for the diagnosis of heart disease. Optimal parameters have been tried to improve the classification success of the algorithm. The first parameter for the KNN algorithm is the distance method, and Manhattan, Euclidean, Chebyshev and Cosine measurements are preferred for this parameter. The other parameter is the number of neighbors and the only numbers between 1 and 15 are tested to identify the most suitable number of neighbors. The KNN algorithm we use to classify heart disease is coded and run in C++ programming language. In the model evaluation phase, the UCI Statlog (Heart) data was used and the results were obtained based on accuracy and ROC analysis. The highest classification accuracy obtained by the KNN algorithm is measured as 100% and the highest AUC value as 1,00. This value is obtained if the Chebyshev distance measurement and the neighbor’s number is 7.

Keywords: