Breast Cancer Prediction Using Different Classification Algorithms with Various Feature Selection Strategies

• Published in: 2021 5th International Conference on Informatics and Computational Sciences (ICICoS) pp. 18 - 23
• Main Authors: Sabha, Mohamad, Tugrul, Bulent
• Format: Conference Proceeding
• Language: English
• Published: IEEE 24.11.2021
• Subjects: Breast cancer; Classification; Data Mining; Feature extraction; Feature Selection; Information filters; Machine learning algorithms; Malignant tumors; Predictive models; Support vector machines
• ISSN: 2767-7087
• DOI: 10.1109/ICICoS53627.2021.9651867

Breast cancer has always been one of the most dangerous diseases that threaten women's lives. If the disease is not detected in the early stages, it can result in the death of the patient. The term breast cancer is referring to a malignant tumor that happened due to the unexpected development of breast's cells which can probably have the ability to spread through other different parts of the patient's body. The occurrence of cancer is often a result of the abnormal growth of cells in our bodies. Cancers generally are classified into two types, Benign (non-cancerous cell) and Malignant (cancerous cell). The earlier the cancer is diagnosed, the better the patient's chance of recovery. Being able to accurately predict breast cancer present in patients has always been an important issue for cancer researchers. Machine Learning (ML) and Data Mining (DM) have always been a point of interest in the scientific community in the hope that they can yield accurate results. We aimed in this study to predict the tumor at early stages using some classification algorithms. After the dataset was collected and the outliers and skewness in the data set were removed, different classification algorithms were applied, focusing on the effect of the feature selection step in the model building phase. After conducting multiple experiments, we got the best overall accuracy by Support Vector Machine (SVM) classifier based on feature selection using Recursive Feature Elimination (RFE) with Random Forest (RF) technique with an accuracy of 98.25%.