A COMPARATIVE EVALUATION OF MACHINE AND DEEP LEARNING ALGORITHMS FOR BREAST CANCER DIAGNOSIS
Breast cancer still ranks amongst the leading causes of death in woman worldwide and early diagnosis is important to increase the chance of survival and effective treatment. Current milestones in ML & DL have laid down strong platforms for medical diagnosis particularly for cancer diagnosis. The objective of this work is to examine and benchmark different kinds of ML and DL models for early detection of breast cancer using WBCD. Normalization of the dataset is performed, as well as an imputation of missing values, feature selection, correlation analysis with the use of heatmap visualization. A number of algorithms are used, namely Logistic Regression, Support Vector Machines (SVM) kernel: linear, radial basis function, and polynomial, K-Nearest Neighbours (KNN), Naive Bayes, Decision Trees, Random Forest, AdaBoost, XGBoost, CatBoost, Convolutional Neural Networks (CNN) and Artificial Neural Networks (ANN). Furthermore, the blended models containing KNN and SVM with Random Forest are reserved to foster prediction accuracy. To perform the hyperparameter optimization, use of Grid Search is made. For model measurement, commonly used indicators include accuracy, precision, recall, specificity, sensitivity, and F1 score are adopted. The highest accuracy of 98.57% is recorded when the model is trained on 90% of the total dataset. The outcomes also suggest that deeper learning and ensemble methods are superior to conventional recipes of applying machine learning algorithms in early diagnoses and treatment of breast cancer.
Artificial Intelligence, Machine and Deep Learning Algorithms, Data Collection and Analysis, Evaluation, Confusion Matrix, Visualization.