Özet:

In the presence of variables that are difficult to model as a result of applications on different data sets, it is recommended to use machine learning regression methods as an alternative to the classic regression methods. Health spending is a variable that is difficult to model, and in literature machine learning has not been found in a study in which this variable is modeled by comparing the methods of regression. In this study, a multi-regression model was created in order to predict the health expenditure per person. The results of the performance of the Lasso Regression, which is achieved when the different hyperparameter values are determined, are compared with the support vector machine Regression with the random tree Regression. The study was based on the value of lamda (λ) for Lasso Regression as the hyperparameter value, the number of trees for random tree Regression, the value of epsilon () for support vector Regression. When the results are performed by the application of cross validity of the "k" piece ranging from 5 to 50, it was found that the performance results of the machine learning regression methods showed statistically significant differences in terms of R2, RMSE and MAE values (p<0.001). When the statistical test results were studied with the surface and strip graphs of the forecast performance, it was determined that random tree regression (R2 0.7500, RMSE ≤ 0.6000 and MAE ≤ 0.4000) had better forecast results according to different hyperparameters. The results of the study are expected to contribute to the determination of the most appropriate hyperparameter values when machine learning regression methods are used for the studies in which health expenditure is modeled.

Anahtar Kelimeler: