The use of machine learning (ML) in the analysis of microbiome data is becoming more common due to the ability to process high dimensional data with small number of samples [8]. There are several pipelines for microbiome data analysis that allow researchers to perform statistical analysis to identify representative sequences related to a disease. Once these sequences are identified, researchers apply ML techniques to differentiate between groups, e.g. between healthy controls and patients [6]. Nevertheless, it has been shown that working with a small number of samples can cause overfitting. A proven solution to avoid overfitting is to use nested cross-validation [8] to produce robust and unbiased results regarding the number of samples. In this work, we propose a novel pipeline that combines DADA2 [2], which is an open-source package for modeling and correcting errors in Illumina-sequenced amplicon inferring sample sequences, and a Recursive Ensemble Feature Selection (REFS) [3,4], method to discover biomarkers, applied to 16s rRNA sequencing gut microbiome dataset PRJEB33711 [1]. The aim of the selected study [1] is to show that the gut microbiome of Inflammatory Bowel Disease (IBD) patients is less diverse compared to healthy individuals.