AI-Assisted Prognostic Classification in Breast Cancer Using Deep Neural Networks

Breast cancer is a significant health issue facing the world, and proper prognosis prediction is vital for effective treatment planning and patient management. Traditional statistical tools are not always capable of capturing nonlinear relationships between clinical and pathological variables in terms of their impact on survival. In order to eliminate this drawback, this paper proposes an AI-based method that classifies prognostics in breast cancer using Deep Neural Networks (DNN). The dataset was preprocessed to eliminate duplicates and missing values, which is a publicly available dataset provided by Mendeley Data comprising$\text{1 5, 0 5 4}$patient records, and it produced$\text{2, 4 8 1}$complete samples. The data was encoded, and categorical attributes were standardized, after which the data was stratified and divided into training and testing sets in an$80: 20$ratio by stratified sampling. A custom DNN architecture was developed and optimized through ablation studies evaluating activation functions, normalization layers, dropout, regularization, optimizers, learning rates, and batch sizes. The evaluation of the model was done through 5-fold cross-validation, confusion matrix analysis, and ROC-AUC curves, as well as comparative evaluation with the traditional machine learning models. The suggested model, which combines LeakyReLU activation and Batch Normalization, regular analysis performed better with an accuracy of$0.9786 \pm 0.0004$, precision of$0.9748 \pm 0.0005$, recall of$0.9812 \pm 0.0005$, F1 score of$0.9780 \pm$0.0004, specificity of$0.9885 \pm 0.0002$, MCC of 0.9678, and Kappa of 0.9673. These results support the fact that the suggested AI-driven DNN framework is more efficient than traditional models, as it is a powerful and trustworthy tool that may be used to predict data-driven prognosis in breast cancer.