Enhanced Breast Cancer Classification using Ensemble Learning

One of the most common cancers that affect women worldwide is Breast Cancer (BC), early and precise identification is essential for improvement in treatment results and survival rates. This study employs an ensemble deep learning strategy to classify breast cancer using the BreaKHis dataset. In order to ensure data integrity and resolve any potential inconsistencies, the dataset is preprocessed. For an efficient model assessment, the data is divided into training and testing sets. In order to capture a variety of feature representations, the model architecture combines the outputs of VGG16, ResNet50, and InceptionResNetV2 feature extractors through a concatenation layer using the feature-level stacking technique. To increase accuracy and improve the classifying process, more dense layers are added. To avoid overfitting and improve convergence, the ensemble model is assembled with binary cross-entropy loss and optimized using the Adam optimizer. The ensemble model’s potential to enhance breast cancer detection and provide more trustworthy medical evaluations is clearly demonstrated by the use of evaluation measures including ROC-AUC, accuracy, precision, recall, F1score and confusion matrix to evaluate model performance.