Prediction and Classification of Breast Cancer Depending on Risk Factors Using Data Mining Techniques: Case Study Palestine رسالة ماجستير

Abstract

Breast cancer is the most frequent cancer in women worldwide. Because of increased life expectancy, the incidence of breast cancer is rising in developing countries. Breast cancer (BC) prediction and classification are important issues to detect cancer in the early stages. There are many risk factors for breast cancer. Due to global environmental problems, the way of life, and the increase in the world population, it is possible that in the coming years the number of people affected will increase. The majority of the women do not go to perform recurring screening mammograms. This study used data mining techniques to predict and classify BC disease, and this is applied by Palestinian patients’ datasets containing the diagnosis of cases (malignant, healthy and benign) and risk factors. The dataset contains 1794 records; 635 malignant breast cancer cases, 570 healthy controls and 589 benign cases. Breast cancer cases were collected from mammography electronic Registry system in MOH Directorate. Each record has 30 features (Variables) plus the class attributes. The class is formed as healthy, malignant, or benign cases. Multiple data mining algorithms was used such as neural networks, boosted decision Tree, Random Forest and Support-Vector Machine. The performance of models was choosing by compared the data mining algorithms with each other. The research evaluated and compared the performance of different machine learning (ML) algorithms in predicting breast cancer among Palestinian women and choose the best ML algorithm to develop a breast cancer prediction model. vi The result shows that the best four models were developed for prediction the breast cancer. The first model for detection BC for three class (malignant, benign and healthy). The results of this model were 96.7% accuracy, 95.2% Precision, 95.1%Recall, 95.2% F1 Score and 97% AUC. While the performance of second models for prediction the BC (Discrimination between malignant and healthy cases) was 98.5% Accuracy, 97.8% Precision, 99.3% Recall which meaning very low false negative, 98.6% F1 Score and 99.9% AUC. The third model was developed to detection the BC (Malignant and benign) has performance with 95.0% Accuracy, 98.0% Recall and 99.5 % AUC. Finally, the model for predicting of BC (healthy and benign class) was developed with 89.4% accuracy, 90.9% Precision, 93.9% Recall which meaning very low false negative, 92.4% F1 Score and 96.1% AUC. The models can be used for diagnosis and to assist doctors in determining whether or not a large number of women require early detection via mammography or other means. This allows women who are suspected of having breast cancer to be given priority for appointments at breast cancer clinics.