Random Subspace Ensemble Learning for Cancer Detection Based on Microarray Data

Microarray technology has fast developed to serve a massive amount of data in the medical field. Microarray data are critical in cancer research for identifying and classifying almost all forms of cancer tissue. Unfortunately, one of the challenges in processing microarray data is the problem of high dimensionality, so that ordinary machine learnings are not sufficient to be used. Therefore, in this research, we introduce a way to overcome this challenge by using the Random Subspace (RS) ensemble learning method. We compared five different machine learning algorithms as basis classifiers, i.e., logistic regression (LR), decision tree (DT), support vector machine (SVM), artificial neural network (ANN), and Bayesian logistic regression (BLR), in terms of their classification and receiver operating characteristic (ROC) performances. We performed these techniques on four microarray datasets from cancer studies such as colon, leukaemia, breast, and lung cancer datasets. Research findings revealed that the RS ensemble on the basis of DT produced perfect performance than the others. Meantime, LR and BLR were great alternatives too. Both of them showed excellent results and became the second-best classifiers for cancer detection. Finally, the RS ensemble on the basis of ANN has shown the worst results on all cancer datasets.