A Hybrid PSO-GWO Optimized Neural Network with LIME-Based Explainability for Robust Breast Cancer Diagnosis

Breast cancer is one of the main causes of death among women in many countries of the world and this is the reason why better and interpretable diagnostic tools are required. We present a new hybrid optimization method that combines Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO) in this paper purposed to optimize neural network hyperparameters such as neural network hidden units and learning rate, among others. The hybrid PSO-GWO algorithm combines the global exploration and local exploitation ability and achieves better classification accuracy on Wisconsin Breast Cancer dataset. The optimized neural network had good accuracy (about 96%) and AUC 0.99 showing a good discrimination between benign and malignant cases. Moreover, to obtain feature-level explanations and aid clinical decision-making procedure, localizable explanations are obtained through LIME (Local Interpretable Model-Agnostic Explanations). Conducting comparative analysis with the traditional classifiers (Logistic Regression, SVM, Random Forest), it can be noted that the proposed model has competitive or better performance, as well as transparency. The process is completely replicable and scalable on Kaggle, which opens the way to the possibility of more people getting access to it, and subsequent integration into clinical practice