Effective Data Dimensionality Reduction Workflow for High-Dimensional Gene Expression Datasets

While moving towards the era of 'Big Data', the scourge of dimensionality is growing an example of the most concerned obstacles in bioinformatics and biomedical research. Typically, an omics classification involves irrelevant and unnecessary features that can take a long time to compute and reduce classification performance. Previously, various researches showed that combining univariate and multivariate feature selection methods may enhance the enforcement of classification. In this research, we have proposed a workflow that can provide better classification performance by using fewer variables for gene expression data. To establish our statement, we started by taking four gene expression datasets: GSE5325, GSE6919/GPL8300, GSE6919/GPL92, and GSE6919/GPL93. We applied Student's t-test to discard redundant features. After that, Principal Component Analysis (PCA) was exercised to reduce the dimension of data. Wrapper Recursive Feature Elimination (RFE) method was performed over the reduced data to obtain the best combination of PCAs for better performance. Finally, the Support Vector Machine (SVM) was utilized to measure performance, and outcomes were compared with the previous researches. The results showed that our proposed approach produced a better performance with much fewer variables for gene expression data. All our research resources, documents, programs and snippets are located at https://github.com/Srizon143005/DataReductionWorkflow.