Explainable Machine Learning-Based Overall Survival Classification in Prostate Adenocarcinoma Using Integrated
Hasan Anıl Kurt1, Sabire Kılıçarslan2, Merve Meliha Çiçekliyurt3
1Department of Urology, Faculty of Medicine, Çanakkale Onsekiz Mart University, 17020 Çanakkale, Turkey.
Diagnostics (Basel, Switzerland)
|May 13, 2026
Summary
An explainable machine learning model accurately predicts prostate cancer survival using clinical and molecular data. This approach improves prognostic accuracy beyond traditional methods for better patient stratification.
Area of Science:
- Oncology
- Bioinformatics
- Machine Learning
Background:
- Prostate adenocarcinoma shows significant patient heterogeneity, challenging current prognostic tools.
- Prostate-specific antigen (PSA) assessment alone is insufficient for reliable survival prediction.
- There is a need for data-driven approaches integrating multi-dimensional data for improved outcome stratification.
Purpose of the Study:
- To develop and evaluate an explainable machine learning framework for predicting overall survival in prostate adenocarcinoma.
- To leverage multi-dimensional clinical and molecular data for enhanced prognostic accuracy.
Main Methods:
- A machine learning pipeline utilized clinical and laboratory data from 494 TCGA PanCancer Atlas patients.
- 16 clinically relevant features were selected; missing values were imputed, and class imbalance was addressed using SMOTE.
- A hybrid Gradient Boosting Machine and random forest (GBM + RF) ensemble model was evaluated alongside other classifiers using 10-fold cross-validation.
Main Results:
- The hybrid GBM + RF model achieved 97% accuracy and 0.95 ROC-AUC, outperforming single classifiers.
- Ensemble models effectively handled missing data and class imbalance.
- Key survival predictors included progression-free survival, hypoxia scores, genomic instability, and immune variables, validated by Cox regression.
Conclusions:
- An explainable ensemble machine learning approach accurately predicts overall survival in prostate adenocarcinoma.
- The framework offers a robust foundation for precision urology decision-support systems.
- Further validation in independent cohorts is warranted.
