Application of Machine Learning Algorithms for Predicting Survival After Pediatric Bone Marrow Transplantation

Creative Commons License

Degirmenci Akdeniz M., Dogan T. U., Isler Y.

Journal of Artificial Intelligence with Applications, cilt.7, sa.1, ss.19-28, 2026 (Hakemli Dergi)

Özet

Cancer is a major health problem affecting all age groups worldwide, including children. Bone Marrow Transplantation is widely used in the treatment of some types of cancer originating from blood and bone marrow. However, the risk of post-transplant mortality is high due to complications such as infections, toxicity, and graft-versus-host disease. Therefore, post-transplant survival prediction plays a critical role in optimizing treatment decisions and improving patient outcomes. In this study, different machine learning algorithms and feature selection methods were applied to predict the survival status of children after Bone Marrow Transplantation treatment using a publicly available dataset provided by the University of California, Irvine Machine Learning Repository. The original dataset contained 187 samples and 36 features. After data preprocessing, it was reduced to 182 samples and 34 features before being used for the experimental analysis. 6 distinct classifiers such as Support Vector Machines, Decision Tree, Logistic Regression, Naive Bayes, Ensemble Learning, and Multi-Layer Perceptron, were tested for prediction. 3 feature reduction methods were used for feature selection: Principal Component Analysis, Chi-Square Test, and One-Way Variance Analysis. Among these classifiers, the Support Vector Machines yielded a maximum of 81.87% accuracy using both 16 features selected by the Chi-Square Test method and 10 features selected by the One-Way Variance Analysis method. On the other hand, the commonly applied One-Way Variance Analysis method resulted in a maximum of 80.77% accuracy using a Decision Tree classifier with 31 features. In conclusion, although Principal Component Analysis is a widely used feature reduction method, Chi-Square Test and One-Way ANOVA may provide higher classification performance with fewer features for survival prediction in children undergoing Bone Marrow Transplantation.