Hybrid Selection of Breast Cancer Risk Factors in Cuban Patients

Breast cancer is a worldwide public health problem, a disease that, although its risk factors are recognized by international health institutions, may vary from region to region. Knowing which risk factors are relevant for a certain type of population is one of the main challenges for early diagnosis and prevention of the disease. In this work, a hybrid selection of features was performed by integrating the voting of filter and wrapper techniques on a set of risk factors of 1,697 Cuban patients. With the Pareto principle on the voting of the different feature selection techniques, 8 out of 15 features were selected: number of biopsies, family history of breast cancer, tobacco use, breastfeeding, number of children, atypical hyperplasia, alcohol consumption, and obesity. The precision with which the features were selected reached an AUC of 0.999 and 99.2% accuracy. The experimental results showed that the identified risk factors have a significant effect, which can be used to develop more accurate breast cancer prediction models and improve prevention and early diagnosis in Cuban patients.