Bulanık Çıkarsama Sistemi ile Makine Öğrenmesi Yöntemlerinin Sınıflandırma Performansını Karşılaştıran Bir Simülasyon Çalışması: Metodolojik Araştırma

Abstract

Objective: Classification is one of the most important research topics of machine learning that aims to correctly predict the target class for each case in the data. In this study, classification performances of fuzzy inference systems that learn from experts and machine learning methods that learn from the data were compared in different sample sizes. Material and Methods: This study was planned as a methodological research. The machine learning algorithms used in the comparison are Multilayer Perceptron, Random Forest, Support Vector Machine, which are frequently encountered classifiers in the literature. The dataset were generated for 6 (sex, chest pain type, max heart rate, exercise induced, oldpeak and major vessels) independent variables determined by variable importance, preserving the characteristics of heart disease data in the University of California Irvine database. Sample sizes were determined in four different sizes as 100, 250, 500 and 1,000. The datasets were divided into two separate sets randomly, with 70% training set and 30% test set. In fuzzy inference systems, the fuzzy rules were automatically generated and Chi's technique was used to create the rules. Accuracy, precision, sensitivity and Fmeasure were used as performance metrics for classification to compare four methods. Results: As a result of the study, it had been observed that fuzzy inference systems are affected by the sample size, and the classification performance is better than other methods as the sample size increases. Conclusion: In general, it has been observed that as the sample size increased, the classification performance of the methods increased.