Breast cancer detection with logistic regression improved by features constructed by Kaizen programming in a hybrid approach

Breast cancer is known as the second largest cause of cancer deaths among women, but thankfully it can be cured if diagnosed early. There have been many investigations on methods to improve the accuracy of the diagnostic, and Machine Learning (ML) and Evolutionary Computation (EC) tools are among the most successfully employed modern methods. On the other hand, Logistic Regression (LR), a traditional and popular statistical method for classification, is not commonly used by computer scientists as those modern methods usually outperform it. Here we show that LR can achieve results that are similar to those of ML and EC methods and can even outperform them when useful knowledge is discovered in the dataset. In this paper, we employ the recently proposed Kaizen Programming (KP) approach with LR to construct high-quality nonlinear combinations of the original features resulting in new sets of features. Experimental analysis indicates that the new sets provide significantly better predictive accuracy than the original ones. When compared to related work from the literature, it is shown that the proposed approach is competitive and a promising method for automatic feature construction.