An Application of the Feature Selection Method Based on Pairwise Correlation for Diagnosis of Ovarian Cancer with Machine Learning

Many machine learning classification problems have high dimensions, and efficient and effective feature selection algorithms are needed to determine the relatively essential features in the dataset. Gene data is often preferred in feature selection applications because it contains many features due to its structure. In addition, it is known from studies in the literature that gene selection plays a significant role in cancer detection. One of the cancer types with very high treatment success in the early period is ovarian cancer. For this purpose, it was aimed to select genes with high descriptiveness in cancer diagnosis by using the ovarian cancer dataset, which is a publicly available dataset. In this study, the feature selection method based on pairwise correlation, which is very new in the literature, was used for classification. Firstly, a feature selection application was made, and 38 genes with the highest cancer descriptors were determined. Then, the classification process was carried out using eight different classification algorithms. After the classification process, the lowest success was for the Extra Tree classification algorithm (with 96.44% accuracy), while the highest was for the Multi-Layer Perceptron, Stochastic Gradient Descent, Logistic Regression, and Support Vector Machine (with 100% accuracy). Although there are many studies on feature selection in the literature, this study is the first application of the current method. In this sense, it is thought to contribute to the literature.