<![CDATA[Obesity is a major non-communicable disease burden, yet population-level dietary guidance does not capture interindividual metabolic variability related to gut microbiome composition. Machine learning (ML) may support microbiome-informed precision nutrition, but evidence remains heterogeneous across classification, response-prediction, and intervention-adjacent studies. This systematic review evaluated ML models that use gut microbiome data for obesity-relevant classification, dietary response prediction, and personalised nutrition outcomes. A PRISMA 2020-guided search of PubMed/MEDLINE, Scopus, Web of Science, and IEEE Xplore identified English-language peer-reviewed articles published from January 2015 to February 2026. Eligible studies included adult cohorts with overweight/obesity participants, BMI-stratified microbiome data, or dietary-response outcomes relevant to obesity-related precision nutrition. Risk of bias was assessed using PROBAST, Newcastle-Ottawa Scale, and RoB 2.0 as appropriate. Twelve studies met the inclusion criteria and were assigned to dominant evidence families: seven classification or biomarker-modelling studies, two dietary or postprandial-response prediction studies, one small, blinded, personalised-diet validation within a PPGR cohort, one prospective, controlled-feeding cohort, and one longitudinal weight-loss prediction study. XGBoost-based classifiers showed the strongest comparative support for static obesity/BMI classification (peak internal AUC 0.90), with SHAP improving interpretability in the best-performing study. Random Forest/MetAML performance was variable, ranging from cross-study AUCs around 0.60 to internal or robust-signature AUCs of 0.83-0.85. The main translational finding was a generalisation envelope from a peak internal AUC of 0.90 to approximately 0.60 across cross-study/meta-analytic benchmarks. Specificity was not reported in any classification-capable study, and calibration metrics were absent across all included studies. Microbiome-based ML shows early promise for precision-nutrition CDSS, but clinical translation requires external validation, compositional-data-aware preprocessing, complete metric reporting, and prospective dietary trials]]>
