A Combined Feature-Learning Method Based on Simulated Annealing Algorithm and Genetic Programming (Case Study: Malignant Breast Cancer Diagnosis)

Nowadays using machine learning tools in different areas such as disease detection is expanding. Origins of this expansion can be found in human's unstable performance and stable performance of machine learning tools. Criticality of detection in areas such as medical proves the need for improvement in machine learning methods. feature reduction and feature learning are two ways that cause to precision increment. In this paper precision of machine learning algorithms is increased by feature learning. The proposed method contains three steps: data quality increment, feature selection, and feature learning. In the first step missing values are replaced with mean or mode (distribution index). In the second step a simulated annealing-based algorithm is presented to utilized as feature selection process and finding the best subset of features. In the final step, a genetic programming algorithms is presented to do the feature learning step. The proposed method is evaluated on two benchmark datasets (WBCD and WDBC). The results show performance improvement in machine learning algorithms in terms of precision if the proposed method used.