Breast Cancer Dataset from Coimbra: Pre Ratings of Its Value to Machine Learning and Diagnosis

Abstract

This study aimed to evaluate a relatively new dataset developed to facilitate the primary diagnosis of breast cancer, collected by the University Hospital Centre of Coimbra in Portugal. Based on these assessments, the authors sought to develop a clear visual classifier to assist medical professionals in prediction and monitoring. This classifier utilizes routine blood test results along with physical data, offering a more straightforward and costeffective alternative to traditional mammographic studies. The Coimbra Breast Cancer Dataset (CBCD) includes the following attributes: Age, Body Mass Index (BMI), Glucose, Insulin, Homeostatic Model Assessment for Insulin Resistance (HOMA-IR), Leptin, Adiponectin, Resistin, and Monocyte Chemoattractant Protein-1 (MCP1). The visual classifier was designed using Java-based machine learning algorithms within the Java-based WEKA software (version 3.9.6). Its well-designed interface enables clinicians, even those without expertise in machine learning, to use these algorithms effectively. The nine attributes of the CBCD were statistically categorized into three subsets based on their relevance to the overall model. This organization may help reduce the dimensionality of the diagnostic dataset while allowing specific classifiers to exhibit their unique preferences. A properly tuned JRip classifier demonstrated acceptable performance with the entire dataset and was effective in reducing it to six or even four attributes. The primary advantage of this classifier lies in its decision rules, which are easy for medical professionals to interpret and apply.