Enhancing Lung Cancer Survival Prediction: A Comparative Study of Hybrid Cox-SVM and Logistic-SVM Models

The research compares the survival performance of hybrid survival models that merge traditional statistical techniques with machine learning to forecast survival outcomes in lung cancer patients. The study compares a group of 165 patients from Tanta Oncology Institute and Kafr El-Sheikh Chest Hospital between 2020 and 2024 by contrasting individual models—Cox Regression, Logistic Regression, and Support Vector Machines (SVM)—with two hybrid models: Cox-SVM and Logistic-SVM. High-risk predictors such as smoking, occupation, age, treatment modalities, COVID-19 infection, and disease stage were found and modeled through Kaplan-Meier analysis for feature selection. The results show that the Cox-SVM hybrid model gives the best results among all the other models with a classification accuracy of 94.5%, sensitivity of 90%, specificity of 96%, and misclassification rate of 5.45%. The Logistic-SVM hybrid then comes into play with a 90.91% accuracy, and then the individual models (Cox: 80.6%, Logistic: 67.88%, SVM: 81.8%). Hybrid model performance is because Cox's hazard-based model combines SVM's capability to work with non-linear relationships and produce better predictive accuracy and clinical relevance. Despite small sample size limitations and omission of certain variables, these results suggest that hybrid approaches in survival analysis are a valuable resource in personalized medicine. These models must be validated with larger, more diverse datasets and other prognostic factors in future research