Comparison of Predictive Performance of Machine Learning Methods in the Diagnosis of Crimean-Congo Hemorrhagic Fever

Authors : Esra Gültürk, Hüdaverdi Bircan, Erdem Karabulut, Nazif Elaldı

Pages : 52-61

Doi:10.51972/tfsd.1660325

View : 52 | Download : 46

Publication Date : 2025-05-31

Article Type : Research Paper

Abstract :Purpose: This study aims to compare the performance results of the machine learning methods “Support Vector Regression, Random Forest, Regression Tree and Nearest Neighbor Regression models on the dataset of Crimean-Congo Hemorrhagic Fever Diagnosis. Materials and Methods: The data of all patients who were hospitalized in Cumhuriyet University Faculty of Medicine, Infectious Diseases and Pediatrics service with the diagnosis of Crimean-Congo hemorrhagic fever between 2009 and 2011 were taken from the service records. During these three years, 6125 data entries were made for a total of 245 patients. A total of three groups of patient data were used in the study: adult, pediatric and all patients. Each scenario was repeated 1000 times with the Boostrap resampling method and the mentioned regression methods were applied in each repetition. To compare the performance of the regression models, the mean squared error and the percentage of explanatory variables were analyzed. Results: Among the regression methods for the real data set, the regression model with the highest explanatory percentage and the lowest mean squared error was found to be the best performing regression model for all three groups. Conclusion: As a result of the simulation study according to real data and scenario structures, the best prediction regression method was found to be support vector regression.
Keywords : K-En Yakın Komşu, destek vektör regresyonu, rastgele orman, regresyon ağacı, makine öğrenmesi

ORIGINAL ARTICLE URL