Prediction of Breast Cancer Grade Using Explainable Machine Learning

The standards for the survival analysis of breast cancer are Cox Proportional Hazards and Kaplan-Meier Survival model. Machine Learning has generated outputs at least better than classical techniques, but they are ignored due to their lack of explainability and transparency, which are crucial for their acceptance in clinical settings for the prognosis and diagnosis of Breast Cancer. Explainable Machine Learning is a set of methods and strategies that allow a model to comprehend and believe machine learning results. In this study, the performance of the Cox Proportional Hazard classifier and the Extreme Gradient Boosting classifier in predicting the survival analysis is compared using microarray gene expression data from the National Center for Biotechnology Information. Results show that Extreme Gradient Boosting performs better than traditional Cox Proportional Hazard regression (c-index: 50.4), with a c-index of 97.7. The forecasts generated by the models are additionally explained using Shapley Additive explanation values. Specific genes had an impact on both the models’ predictions and their intuitiveness. Finally, Extreme Gradient Boosting can generate explicit knowledge about how a model generates its forecasts, which is crucial in boosting confidence and encouraging the adoption of innovative machine learning techniques for the prognosis and diagnosis of breast cancer.