Metagenomics is an advanced research field that has revolutionized the study of genetic material extracted from environmental samples. It has wide applications, from environmental studies to human health, especially in understanding the gut microbiome. However, artificial intelligence application to disease diagnosis remains complex, requiring explanatory methods to clarify how artificial intelligence makes decisions. In this study, we aim to improve disease prediction by applying explainable Artificial Intelligence (AI) techniques in combination with a robust Random Forest algorithm. The goal is to compare Explainable AI (XAI) methods to identify the most effective one and discover new features associated with disease, considering patient-specific and geographical factors. The study focuses on colorectal cancer classification on four metagenomics datasets. Experimental results and comparisons between different XAI methods show the potential for improving disease prediction accuracy. At the same time, several features affecting colorectal cancer were identified, emphasizing the role of XAI in disease diagnosis and opening up further research directions in this field. The study’s findings improve the efficiency of disease classification and open up new prospects for future metagenomics and colorectal cancer research.