Significant directed walk framework to increase the accuracy of cancer classification using gene expression data

Abstract

Early diagnosis methods in cancer diagnosis studies are making great challenge as they require the involvement of different fields. Deoxyribonucleic acid (DNA) microarray analysis is one of the modern cancer diagnosis techniques used by scientists to measure the gene expression level changes in gene expression data. From the perspective of computing, an algorithm can be developed to identify more difficult cases. Numerous cancer studies have combined different machine learning techniques for the cancer diagnosis. This study is conducted to improve the cancer diagnosis technique, directed random walk (DRW) from the direction of framework. Improved directed random walk framework is proposed with the new introduced sub-algorithms, a larger directed graph and a different classifier. It is named as significant directed walk (SDW). In this study, six gene expression datasets are applied to study the effectiveness of the sub-algorithm, directed graph and classifier in SDW in terms of cancer prediction and cancer classification. Sub-algorithms of SDW can be further divided into data pre-processing phase, specific tuning parameter selection, weight as additional variable, and exclusion of unwanted adjacency matrix. Besides that, SDW also incorporated four directed graphs to study the usability of the directed graph. The best directed graph among the four is chosen to be part of the structure in SDW. The experimental results showed that the combination of SDW with walker network and linear regression is the best among all. SDW is achieves accuracy of 95.03% in average which is higher by 8.97% compare to conventional DRW for all cancer datasets. This study provides a foundation for further studies and research on early diagnosis of cancer with machine learning technique. It is found that these findings would improve the early diagnosis methods of cancer classification.