Sağkalım verileri için makine öğrenmesi yöntemleri

Survival analysis is a the statistical approach methods used in many fields. The key feature that distinguishes this method from other analysis methods is that it can be used when censored observations are present. In the literature, survival analysis can be categorized as traditional survival analysis methods and machine learning based survival analysis methods. The increasing number of data and variables, the existence of censored observations and the fact that traditional methods require some assumptions make it difficult to analyze survival data with traditional methods. In order to cope with this situation, machine learning methods specific to survival data are used. In this thesis, the traditional survival analysis methods such as the Kaplan Meier, Log rank test and the Cox regression are introduced and machine learning based Survival trees, and Random Survival Forests are studied. In application concordance statistics of Random Survival Forests and Cox regression models were obtained by using both real and simulated data in which different censoring rates were tested. Propher graphical representations were given for the survival curves and variable importance metrics when necessary. As demonstrated by the examples provided in this thesis the traditional approaches are adventageous when assumptions are met and statistical power is high however machine learning based methods work better when censoring is high an assumptions are not satisfied.