Microarray based cancer classification using ensemble learning

Abstract

Cancer diagnosis and classifications are one of the crucial emerging clinical applications of microarray data. The technology of DNA microarray has allowed us to find the gene expression levels of thousands of genes at the same time giving us a huge opportunity for cancer diagnosis and classification. To determine a computational model, from the given data so that the classes of unknown samples are determined, is the primary objective of microarray data. The majority of researches done in this field dealt with gene expression sequence to differentiate between the cancerous gene expression sequence and healthy gene expression sequence. We are using a microarray dataset which contains the RNA sequence gene expression levels of thousands of genes of hundreds of samples. Our dataset contains information about five types of cancer referred as breast, colon, kidney, prostate and lung cancer. We are using five classification models namely Support Vector Machine(SVM), Random Forest, XGBoost, K Nearest Neighbour(KNN) and Deep Learning for classifying the five different cancers and labeling the groups using the labeled data . Using the ensemble learning technique we combined all the five models in order to make an optimal classification model that can differentiate each sample of the test dataset to its respective cancer class. We used t-SNE (t-distributed stochastic neighbor embedding) model for data visualization and creation of five clusters of each cancer. By training and testing our dataset we were able to determine the accuracy of each model. Using our well-structured model we can easily use any unknown patient's RNA sequence gene expression level to identify if he or she is a patient of any of the five cancers or not.