Using a hybrid BPNN-BC model for breast cancer diagnosis

This study proposes a hybrid two-stage architecture for breast cancer diagnosis using the Wisconsin Breast Cancer Dataset (WBCD). The approach integrates a Backpropagation Neural Network (BPNN) with a Bayesian Classifier (BC). Preprocessing comprises median replacement for outliers and feature normalization. The BPNN is trained with a 70/15/15 % split (train/validation/test) and performs the primary classification. Samples with low confidence in the network’s output are forwarded to the BC for a second-stage decision. In computational experiments, the BPNN achieves 94.6 % test accuracy (best MSE = 0.037611 at epoch 8), while the BC attains 100 % accuracy on the designated subset of hard cases (8 observations). The hybrid BPNN-BC scheme reduces false negatives, improves robustness to outliers, and provides interpretable probabilistic estimates properties that are valuable in clinical decision support. The results suggest that combining machine-learning and statistical classification yields more reliable predictions for breast cancer diagnostics. Future work includes expanding the feature set, conducting external validation on clinical cohorts, and benchmarking against alternative ensemble-based baselines.