Simultaneously feature selection and parameters optimization by teaching–learning and genetic algorithms for diagnosis of breast cancer

Currently, development of early and accurate breast cancer (BC) prediction models using computer-aided tools has proven to be beneficial, which in turn low mortality rate related to this disease. However, feature selection (FS) is a challenging task for the identification and characterization of cancers that increase the susceptibility to common complex multifactorial BC diseases, especially when dealing with clinical treatment. Most of the previous FS techniques does not handle important characterization such as removing irrelevant and/or redundant features separately. According to the past research on FS, several evolutionary algorithms have been proposed to address FS problems, but they have to fail for classifying BC survival types. In order to address before-mentioned issues, numerous hybridized models have been intended for selecting best features in effort to increase the accuracy of breast cancer predictive models. It may be cumbersome to obtain the perfect parameters for optimal performance. To resolve the deficiencies of past diagnostic system, in this paper, hybrid teaching–learning optimization (TLBO) and genetic algorithm (GA)-based is proposed consistent wrapper strategy called TLBOG to improve the reliability of evolutionary algorithms. The aim of using GA here is to tackle slow convergence rate and improve exploitation search capability found by TLBO. Most importantly, goal of our approach is to optimize the parameters of support vector machines to have high accuracy in contrast to other machine learning models and select best features subset simultaneously. From the performance evaluation results, we understand that proposed approach is significantly higher than conventional wrapper techniques in terms of accuracy, sensitivity, precision, and F-measure in the WBCD and WDBC databases.