Use of LDA combined with PLS for classification of lung cancer gene expression data

Reliable and precise classification is essential for successful diagnosis and treatment of cancer. Thus, improvements in cancer classification are increasingly sought. Linear discriminant analysis (LDA) is the most effective method of cancer classification in high-dimensional prediction, but there are drawbacks to tumor classification by a formal method such as LDA. We propose a method for lung cancer gene microarray classification that combines a feature reduction approach, partial least squares (PLS), and discriminate method, LDA, for improving classification performance. The real dataset used related to lung cancer gene expression. After bioinformatics data preprocessing, data reduction and feature selection were carried out using PLS and then LDA was used for classification. The results were validated using the accuracy index and gene ontology analysis. Of the total of more than 50,000 genes, 214 genes were shown to have relevance. The classification accuracy of this method was 94.5% and gene ontology analysis results were good. It can be said that the LDA classifier combined with PLS is powerful method. This method can identify gene relationships warranting further biological investigation.