Unified neural network ensemble for accurate breast cancer classification with engineered features

This study presents the Unified Neural Network Ensemble for Breast Cancer Classification (UNNEBC), a novel ensemble model designed for breast cancer classification. An ensemble learning approach based on Convolutional Neural Networks (CNN) was used to improve the classification performance by utilizing three CNN models with different architectures and two fully connected dense networks. To enrich the dataset and improve classification performance, new features named Radius_Texture_Diff, Feature_Std_Dev, and Feature_Variance have been introduced. Radius_Texture_Diff captures the difference between the radius and texture features, potentially highlighting abnormalities in tumor shape and surface characteristics. Feature_Std_Dev represents the standard deviation of all features, providing a measure of variability within each tumor sample. Feature_Variance quantifies the variance across all features. By incorporating these features alongside the original dataset, we aim to enrich the feature space and enhance the model?s ability to capture complex tumor patterns. The model was evaluated using the UCI breast cancer dataset and achieved an outstanding accuracy of 99.42%. It also showed strong performance in metrics such as specificity and sensitivity. The UNNEBC model tackles significant challenges in breast cancer classification, such as unequal class distribution and data variability. By using ensemble learning and integrating XGBoost as a meta-learner, the model leverages the strengths of individual networks and provides more reliable predictions. This study outperforms existing approaches. It also highlights the importance of feature engineering and ensemble learning in advancing breast cancer diagnosis.