A Cost-Sensitive Centroid-based Differential Evolution Classification Algorithm applied to Cancer Data Sets
Jamil Al‐Sawwa, Simone A. Ludwig · 2019
Nowadays, the collected or generated data for some real-life applications such as in the Medical domain and Intrusion Detection, are typically imbalanced. Imbalanced data sets consist of data where one class-label (minority) includes significantly fewer instances compared to other class labels. The misclassification of the minority class-label could be costly in some circumstances. Therefore, the extraction of valuable information from this kind of data poses a challenge to the scientific community. During the last decades, the researchers proposed a centroid-based classification algorithm using differential evolution (CDE) to solve data classification. However, CDE shows an inefficient performance especially when applied to imbalanced binary data sets. In this paper, we propose a cost-sensitive version of CDE based on a new objective function in order to overcome this drawback. We are using four cancer data sets that are imbalanced namely Breast, Lung, Uterus, and Stomach. Furthermore, we analyzed and investigated the performance of our proposed version of CDE for predicting the survivability of cancer patients compared to the performance of the current variants of CDE. Moreover, we compared the performance of our proposed version of CDE with the performance of five cost-sensitive machine learning algorithms. The experimental results demonstrate that our proposed version of CDE improves the performance of CDE when applied to imbalanced binary data sets. Furthermore, the performance of our proposed CDE algorithm outperformed the performance of the current variants of CDE on all data sets in terms of Area Under Curve and G-mean.