Comparative Evaluation of ANN, Decision Tree, and SVM Models for Breast Cancer Classification Using Cytological Features

Abstract

Breast Cancer (BC) is still one of the deadliest causes of death for women, and timely and precise diagnosis is essential for achieving better clinical results with the use of reliable machine learning-based decision support. All three models, namely the optimized Artificial Neural Network (ANN), Decision Tree (DT) and Support Vector Machine (SVM), are evaluated in relation to their predictive performance in a uniform experimental set-up and the best model was identified to classify BC. The data set used to build and test the three supervised machine learning models is called the BC Wisconsin (Original) data set. Data preprocessing techniques such as missing value imputation, target encoding, feature scaling and stratified data partitioning are used to improve the generalization and predictive ability of the model, respectively, whilst hyperparameter optimisation and cross-validation are adopted to obtain the best model. Results and Discussion showed that the SVM model performed best overall in terms of classification accuracy (96.19%)with a balanced precisionof 94.44%in terms of precision, 94.44%in terms of recall, 94.44%in terms of F1-score and 98.83%in terms of AUC, which means its performance was good in terms of generalisation. The ANN model also showed promising results with a high accuracy of 95.24%and the highest AUC of 99.03%, indicating that the model is suitable for modelling nonlinear relationships. The highest recall (97.22%)was achieved by DT model, indicating that this model had excellent sensitivity in detecting malignant cases, but the low precision (81.40%)led to a relatively high false positive rate. The results suggest that model optimisation and careful choice of the algorithm significantly impact the reliability of diagnosis. From the evaluated models, the optimized SVM proved to be the most robust model with fair performance, and it can be used as an effective decision support tool for the classification of BC.