Breast Cancer Diagnosis System

Breast cancer is one of the most common cancers globally, affecting millions of individuals each year. Accurate diagnosis is significant due to its prevalence and potential impact on individuals’ lives. Developing an effective machine learning-based Breast Cancer Diagnosis System addresses the pressing need for improved diagnostic accuracy and aligns intending to enhance personalized medicine. In this study, we utilized the Wisconsin Breast Cancer Diagnostic (WBCD) dataset, which has 569 instances and 31 features that include measures of cell nuclei characteristics. To identify the most discriminative features, various feature selection algorithms were employed, including Variance and Fisher scores, Wrapper, and the Embedded algorithm. We employed six classifiers: Logistic Regression, Naive Bayes, Support Vector Machines, Decision Trees, Random Forest, and K-nearest Neighbors for breast cancer diagnosis. To evaluate model performance, two distinct train-test split ratios were used: an initial 90/10 split for preliminary model training and testing, followed by a more extensive 70/30 split for detailed evaluation. Additionally, cross-validation was conducted employing 3 and 5 folds, facilitating robust performance assessment across various data partitions. The results of Logistic Regression showed 98.2% accuracy using the Wrapper feature selection method with the 90/10 training set. The same accuracy was achieved with SVM and Random Forest algorithms using the Fisher feature selection model with a 90/10 testing set. Accuracy was improved in all models by including feature selection algorithms, emphasizing the importance of selecting the appropriate algorithm, and features, and employing effective dataset-splitting strategies when optimizing classification tasks.