Hyperparameter Selection for a Breast Cancer Classification Using Machine Learning

Breast cancer is a disease that attacks breast cells in women. One way to assist in diagnosis is to use ultrasound and mammography intelligently using machine learning. There are several algorithms of machine learning that need to determine the selection of hyper-parameters in order to produce a model with the best accuracy. Therefore, this research conducts experiments to determine the best performance of hyperparameters for each machine learning algorithm. The research stages include: data preprocessing, feature extraction, classification, and evaluation. The feature extraction process uses depthwise convolution feature fusion, while the classification process uses five machine learning algorithms, namely SVM, Random Forest, KNN, Decision Tree and Logistic Regression with k-fold cross-validation. The results of testing several hyperparameters of each machine learning algorithm show the best average accuracy. On ultrasound images, the best average accuracy is obtained by the svm algorithm with a ‘linear’ hyperparameter of 84.35%, while on mammography images, there is a random forest algorithm with a min_samples_split-2 hyperparameter of 92.80%. After testing, the best performance was obtained using the logistic regression algorithm with an accuracy validation value of 86.50% on ultrasound images and 95.90% on mammography images. In the ROC and precision recall curve, the best performance is produced by the LR algorithm on ultrasound images with a value of 0.95 and on mammography images, the RF algorithm provides the best value of 0.98. The use of appropriate algorithms and hyperparameters can improve accuracy and precision in disease detection and diagnosis.