Classification and rule generation for colon tumor gene expression data

Microarray genome studies discover the relationship between gene expression profiles and various diseases. This relationship generally introduces valuable quantitative information from genome profiles. The information facilitates drugs and therapeutics development to provide better treatments. In this paper we suggest that the statistical learning algorithm, Support Vector Machine (SVM) is a useful classification technique to classify genome profiles. Performance and usefulness of SVM is verified with colon tumor genome data. A comparison of SVM’s performance is made with another popular decision trees based classification technique C5.0. SVM is found to be superior to C5.0 in performance. However, SVM lacks the rule extraction capability. We extract rules to identify the responsible tissues for colon tumor using C5.0. The rules could be used with SVM to reduce the size of microarrays in future.