Theophilus Gyedu Baidoo and Hansapani8 Both survival-specific and machine learning models were evaluated using performance metrics such as Concordance Index (C-Index), Integrated Brier Score (IBS), and Area Under the Curve (AUC). The COX proportional hazards (CPH) model, random survival forests (RSF), and Deepsurv showed strong performance, with RSF achieving a C index of 0.72. Both COX and RSF recorded a minimum IBS value of 0.08. However, machine learning models such as Random Forest (AUC 0.74) and XGBoost (AUC 0.69) showed moderate identification, but lacked a mechanism to process censorship data, a key limitation of survival analysis. In related studies, the authors applied five machine learning classifiers using 13 selected features. LightGBM is optimized through this structured Parzen estimator, achieving 99.86% accuracy, 100% accuracy, 99.60% recall, indicating high potential in highlighting malignant and benign tumors with minimal human intervention.9.
Jialong Xiao, Miao MO, and others10 We compared COX models and machine learning algorithms to predict overall survival in a large breast cancer cohort of 22,176 patients. Their findings revealed that the RSF is slightly above the COX model at a C index of 0.827 compared to 0.814. This highlights the usefulness of RSF in prognostic modeling. Another study investigated modified Weibull distributions that could model the shapes of different hazard rates, including increased, decreased, constant, or bathtub-shaped patterns, and the results are closely aligned with Kaplan-Meier Survival Curves.11. Another study by Tizi and Abdelaziz Berrado12 We compared traditional statistical methods and machine learning techniques for predicting cancer survival. In this study, we used C indexes to compare performance to evaluate models that include random survival forests and Cox regression with Ridge Remulization. The results showed that COX regression struggles with high-dimensional data, but both approaches are performed similarly. Another study applied a machine learning model to predict invasive disease events in 145 patients, indicating that random survival forests with gradient boost outweighed the COX model (C-Index, 0.68 vs. 0.57). These findings suggest that clinical data alone can improve prediction accuracy and reduce the need for costly genetic testing.13.
Surbhu Gupta and Manoj K. Gupta14 To predict postoperative survival of breast cancer, we evaluated a deep learning model that includes a restricted Boltzmann machine (RBM). Using cross-validation, RBM achieved the highest accuracy (0.97) and reinforced the need for continuous evaluation of deep learning architectures for optimal predictive performance.
Research by Sahar A. and El Rahman15 We investigated early breast cancer detection using machine learning algorithms and feature selection on four datasets. Classifier performance varied across datasets: Random Forest with genetic algorithm achieved 96.82% in WBC, C-SVM with RBF kernel reached 99.04% in WDBC, Random Forest with recursive function removed achieved 74.13% in WPBC and decision tree at 83.74%. Another comparative study16 The reported SVM and LDA achieved 93% accuracy, 98% random forest, and 86% logistic regression, showing consistent effects across the model.
Gunjan et al.17highlighting the importance of early breast cancer detection and reviewing advances in AI-based computer-aided diagnostic (CAD) systems. They compared machine learning and deep learning approaches to traditional methods and discussed the benefits, limitations and future directions for medical image analysis. Nermin Abdelhakim Othman et al.18 We proposed a hybrid deep learning model to predict breast cancer survival using multi-omics data from the metablick dataset. This framework combines convolutional neural network CNN-based feature extraction with long-term short-term memory (LSTM) and gate recurrence unit (GRU) classifiers to achieve 98.0% accuracy through fusion of decision-making levels. This model significantly improves survival prediction of a single modality approach and provides a more robust and accurate tool for personalized breast cancer prognosis.
Another study using the Wisconsin Breast Cancer Dataset19 Several classifiers were evaluated, including SVM, K-nearest Neighbors, Random Forest, and logistic regression. SVM is the most accurate, achieving 95% accuracy, reaffirming the role of CAD systems in early detection. Individual comparison of linear and nonlinear models20 Although SVMs are more sensitive, artificial neural networks improve overall diagnostic performance and emphasize the value of nonlinear models of complex datasets.
Research using surveillance, epidemiology, and final results (SEER) data from 2010 to 2019twenty one We have developed an Xgboost model to predict survival in patients with bone metastatic breast cancer (BMBC). This model achieved an AUC score of over 0.79. Prognostic factors such as delayed treatment and income levels were important, and neoadjuvant chemotherapy and surgery to improve outcomes for certain subgroups were improved.
Jain et al.twenty two The aim is to identify the best machine learning model for automated breast cancer diagnosis using the Wisconsin data set. Their results showed that hyperparameter-tuned models such as Xgboost and boost algorithms consistently achieved high accuracy for both benign and malignant classifications. Studies using cancer genome atlas – Breast invasive cancer (TCGA-BRCA) data settwenty three We investigated a multimodal machine learning system for survival prediction by integrating six biomedical modalities. Dimension reduction technology and classifiers (SVM, Random Forest) have improved accuracy and robustness. However, these models do not have a prospective validation of the primary dataset, indicating the need for actual testing.
Yinan Huang, Jieni Li, Mai Li, and Rajender rtwenty four. We reviewed 28 studies applying machine learning models to real-world healthcare data for results up to the event. Random Survival Forests and Neural Networks are commonly used in oncology. This review focused on the underestimation of ML for predicting treatment and highlighted the need for methodological advances to enhance clinical utility.
Research by Chirag Nagpal, Xinyu Li, and Artur Dubrawskitwenty five To avoid the proportional hazard assumption of COX models, we proposed a complete parametric deep learning approach for predictions up to time to event. Their model accurately estimated the survival risk of datasets with complex censorship and competing risks, providing significant advances in parametric survival modeling. M. Darshan Teja and G. Mokesh Rail26 Eight machine learning models for cardiovascular disease prediction were evaluated using University of California Irvine data. Ensemble methods such as random forests and bagged trees achieved the highest accuracy and ROC-AUC. K-fold validation confirmed the reliability of the model and highlighted the effectiveness of the ensemble technique in predictive tasks.
Keren Evangeline I., SP Angeline Kirubha, and J. Glory Precious27 We used the metabric dataset to identify predictors of breast cancer. They compared the Cox Proportional Hazards (COXPH) model, RSF, and DeepHit. The RSF and deep hits outweigh COXPH, both achieved a C index of 0.86 compared to COXPH's 0.85. Key predictors included recurrence-free status (RSF), age of diagnosis, estrogen and progesterone receptor status, and tumor stage (COX proportional hazards), which aided clinical decision-making. Recent research also focuses on enhancing survival predictions through frail modeling28. Another study29patients in non-surgical positions were engineers and para-specialists located at the intersection of manual and non-surgical, and revealed a better survival rate (hazard ratio <0.85).
the study30 Machine learning was employed to predict survival using tumor-related clinical features such as stage, size, and age. Kernel Ridge Regression, K-Nearest Neighbors, Lasso, and Decision Tree models showed high prediction accuracy with effective data integration techniques. Finally, a study using data from the University of Ilorin Teaching Hospital31 Several machine learning algorithms were applied to predict breast cancer survival. Adaboost surpassed other models, achieving 98.3% accuracy and 99.9 AUC, confirming its potential for clinical applications.
Survival analysis is widely used in breast cancer research, but has not been much studied in the context of invasive lobular cancer (ILC). Existing literature generally employs Cox proportional hazards models and random survival forests, with fewer studies examining the performance of other established parametric models, such as Weibull, exponential, logistic, logarithmic logistic, Gaussian, and logarithmic Gaussian distributions. Furthermore, the application of formal model selection criteria such as the AKAIKE Information Criteria (AIC) and Bayesian Information Criteria (BIC) is less common in studies that include machine learning approaches. Therefore, further investigation of diverse modeling techniques and assessment metrics may contribute to a more comprehensive understanding of survival predictions. The purpose of this study is to address this need by comparing multiple parametric and machine learning models of ILC survival prediction to support model evaluation and interpretability in clinically meaningful contexts using AIC/BIC and performance metrics. The objectives of this study were as follows:
-
1.
To investigate the prognostic significance of clinical and pathological factors such as age, tumor grade, AJCC stage, and treatment in breast cancer survival outcomes.
-
2.
Comparative evaluations of parametric survival models and machine learning algorithms are conducted when predicting patient survival using statistical criteria, including AIC, BIC, and ROC-based measurements.
-
3.
To identify the most appropriate predictive model, we evaluated the trade-off between model interpretability and prediction accuracy across various machine learning methods.
