Breast cancer diagnosis using a new evolutionary forest approach

Abstract

Breast cancer remains one of the leading causes of cancer-related mortality worldwide, and early, accurate diagnosis is crucial for improving patient outcomes. Traditional diagnostic methods often rely on manual feature selection, which can be biased and suboptimal, especially in high-dimensional medical datasets. To address this, we propose an automated feature selection framework based on an evolutionary algorithm integrated with a Random Forest classifier. Our method iteratively evolves compact, high-performing subsets of features by maximizing classification performance during cross-validation. This approach eliminates reliance on domain-specific heuristics and improves model generalization. Experimental results on the Wisconsin diagnostic breast cancer dataset demonstrate that the proposed evolutionary random forest model achieves high diagnostic performance, with an accuracy of 98.59%, balanced accuracy of 98.21%, precision of 98.07%, recall of 99.72%, F1-score of 98.89%, and AUC of 99.68% using 5-fold cross-validation. These findings confirm the effectiveness and practicality of the ERF method for breast cancer diagnosis. The approach is fully automated and adaptable to other structured medical classification tasks.