Feature Selection Based on Ranking Metagenomic Relative Abundance for Inflammatory Bowel Disease Prediction

Inflammatory bowel diseases can be severe, but with access to metagenomic data, we can diagnose them and take the necessary steps to prevent further complications. The key to identifying the composition in the human body that causes the disease is carefully selecting features from the metagenomic data. Our research has demonstrated that using the Random Forest machine learning technique to rank the relative abundance of features for disease prediction tasks is reliable. We have also discovered that selecting features ranging from 1 to 50 improves the accuracy of diagnosis. In addition, we have performed an intersection on the Top 10, 20, 30, 40, and 50 features to determine which ones appear in all datasets. Our experiments on six inflammatory bowel disease-related datasets have yielded better results than previous studies.