Özet:

Diabetes is one of the increasing, common health problems worldwide. Diabetes, a chronic disease that is not controlled, can cause destruction and death in many organs, such as eyes, heart, kidneys. Early diagnosis of diabetes is important in order to prevent complications that may occur and improve the quality of life. The machine learning techniques commonly used in the medical field take the role of a smart decision support system for specialists in the diagnosis of different diseases. This study includes classification studies carried out on the PIMA diabetes data set with 6 different machine learning techniques for early diagnosis of diabetes. The main objective in classwork studies is to increase the accuracy of predictions. In this study, 14 different sampling methods were used on the data set to increase the success of classifiers. 90 classification processes have been carried out without sampling and re-sampleing for each machine learning model. The success of each classification process is by 5 different performance meters. The most successful result, with 96,296% accuracy, was obtained in the classification process in which the Rastgele Forest model was used along with InstanceHardnessThreshold's low sampling technique. It has been shown that sampling techniques generally enhance the success of classifiers and give more successful results when used together with collective learning methods. On the same set of data in literature, compared to the latest studies conducted using various machine learning methods, the success achieved in this study has been higher than others.

Anahtar Kelimeler: