Optimized artificial neural networks for breast cancer diagnosis prediction

Breast cancer is the second leading cause of cancer mortality in females. A significant amount of research work has been carried out in breast cancer detection. Recent studies have proposed that early detection of breast cancer helps to improve the survival rate. The current research aims to detect Breast Cancer using Artificial Intelligence-based prediction models. In this research, we have explored the performance of the ANN model and highlighted the significance of hyperparameter Optimization of the ANN model. The model has been Optimized using three renowned optimization methods: Bayesian Optimization, Grid search Optimization, and Genetic Optimization on the Wisconsin dataset. The dataset explored in the study is a benchmark dataset publicly available in the UCI and Kaggle repositories. The experimental analysis includes data preprocessing strategies such as label encoding and data standardisation. The results clearly show that Genetic Optimization has attained 96.8% accuracy, 94% AUC-ROC score, 96.9% F1-Score, and 92.87% MCC Score and outperformed Bayesian and Grid-Search Optimization techniques. All the models were validated using 10-fold cross-validation. Also, it is noteworthy that the model’s performance greatly depends on the hyperparameters (learning rate) used to train the model, as the model performed best with a learning rate of 0.1 and the fewest prediction outcomes at a slow learning rate (0.0001). Furthermore, we present 95% confidence intervals for key metrics to validate the robustness of the model. Current research asserts the significance of Optimization techniques for attaining higher prediction outcomes. This research presents an empirical overview and promotes the dominance of AI-driven prediction models in clinical decision-making. This study stresses on deployment of such models in real-world settings.