Dimensionality Reduction on Metagenomic Data with Recursive Feature Elimination

The Fourth Industrial Revolution has a significant impact on many aspects, which help improve and develop significantly. These beneficial works give a better life for all society. When we mention the medical or healthcare field, there has been much creative and vital research that promotes everyone's life. Inflammatory Bowel Disease (IBD) is one of the most dangerous diseases that can cause millions of deaths every year. In this research, we would like to raise a topic about IBD diagnosis using metagenomic data to advance prediction for initial detection. The problem is not well-studied adequately due to the lack of data and information in the past. However, with the rapid development of technology, we obtain massive data where a metagenomic sample can contain thousands of bacterial species. To evaluate which species are essential to the considered disease, this work investigates a dimension reduction approach based on Recursive Feature Elimination combining with Random Forest to provide practical prediction tasks on metagenomic data. The relationship between bacteria causing IBD is what we have to figure out. Our goal is to evaluate whether we can make a more reliable prediction using a precise quantity of features decided by Recursive Feature Elimination (RFE). The proposed method gives positively promising results, which can reach 0.927 in accuracy using thirty selected features and achieve a significant improvement compared to the random feature selection.