Oncogenes Classification Measured by Microarray using Data Mining Computational Methods

Abstract

Introduction: In recent decades, cancer has been given a great dimension, becoming an evident world public health problem. The World Health Organization estimates that, in 2030, 27 million cancer cases and 17 million cancer deaths can be expected. Faced with this alarming scenario, Data Mining brings methods and tools to help building more meaningful knowledge about cancer. Objective: This paper aims to apply five traditional Data Mining methods on the NCI60 dataset. The database was created with data from microarray experiments, with levels of expression of 1,000 genes grouped into nine cancer classes. Method: The methods used on this paper are: J48, Random Forest, PART , IBK and Naive Bayes, which belong to the Weka environment, very traditional in Data Mining. Due to the low number of records for some cancer classes, the validation of the results obtained by the classifiers used the 3-fold cross validation. Results: The classifier which obtained the highest accuracy was IBK, while J48 and PART classifiers drastically reduced the set of genes, building high level knowledge as trees or rules. Conclusion: The results of this study can be used as tools aiming at assisting cancer cure research and may be used in the classification of new cases or to further improve understanding of gene/gene and gene/cancer relations.