Machine Learning Approaches for Detecting and Classifying the Cancer type using Imbalanced Data Downsampling

Abstract

One of the most important applications of medical data mining is the early diagnosis of diseases with high accuracy. In the meantime, timely diagnosis of cancer as one of the main causes of death is of special importance. However, the classification and diagnosis of cancer is challenging due to the unbalanced nature of related data. In the data related to cancer disease, there is usually a minority class (patient samples) and a majority class (healthy people samples), which diagnoses the disease from the minority samples, and this is a challenge for the classifiers. This work investigated the problem of classifying the imbalanced data related to cancer disease using a machine learning approach based on the K-Nearest Neighbor (KNN) clustering technique. In this method, the insignificant samples of the majority class are removed, and the data are balanced. The proposed method is simulated and evaluated on 15 cancer datasets selected from the general SEER database. The simulation results approve a high classification of cancer type based on the average detecting accuracy criterion of more than 90%. Moreover, the current result is more efficient and improves classification accuracy compared to the methods proposed by other researchers in the literature survey.