Evaluation of computational classification methods for discriminating human heart failure etiology based on gene expression data

Human heart failure is a complex syndrome that can be initiated by a variety of clinical conditions and genetic factors. Gene expression profiling offers opportunities to study changes in transcriptional activity in heart failure samples of different etiologies. This paper evaluates machine and statistical learning models for supporting the identification of heart failure etiology based on gene expression data. Six supervised classification models were evaluated on a publicly- available human heart failure dataset. The Naive Bayes, Support Vector Machines, and k-Nearest Neighbours achieved the most significant prediction performances. Using a correlation coefficient-based gene-ranking criterion, the impact of the number of genes on the prediction performance was investigated. Information from the top 5 genes was sufficient to accurately distinguish between ischemic and idiopathic samples.