Comparative Analysis of Machine Learning Algorithms for Breast Cancer Prediction Based on Fine Needle Aspirate Image Features

Breast cancer is the most frequently diagnosed cancer in women, so accurate and explainable prediction models could support data-driven screening. This study compares seven machine learning algorithms, namely Logistic Regression, SVM (RBF), KNN, Decision Tree, Random Forest, Gradient Boosting, and ANN (MLP), for predicting tumour malignancy from 30 cell-nucleus morphometric features of fine needle aspirate images. The dataset contains 569 samples (357 benign, 212 malignant) without missing values or duplicates. The data were split 80:20 with stratification; normalisation and hyperparameter tuning were performed in a pipeline with 5-fold GridSearchCV on the training data only, and model selection used 5 × 3 repeated stratified cross-validation. SVM (RBF) achieved the highest cross-validated F1-score (0.9714 ± 0.0120) and, on 114 test samples, reached an accuracy of 0.9737, precision of 1.0000, recall of 0.9286, F1-score of 0.9630, ROC-AUC of 0.9927, and MCC of 0.9442, with three false negatives and no false positives. Random Forest matched this test accuracy, whereas Logistic Regression achieved the highest ROC-AUC (0.9960). Permutation importance identified texture_worst, concavity_worst, and concavity_mean as the most influential features. Overlapping bootstrap confidence intervals indicate that differences among the best models are not significant, and the model has not been clinically validated.