A Systematic Literature Review of Adaptive Neuro Fuzzy Models for Early Heart Attack Detection

Cardiovascular prediction increasingly combines data-driven learning with interpretable reasoning, yet reported performance is difficult to compare because studies use different datasets, validation schemes, optimization targets and clinical endpoints. This systematic literature review examines adaptive neuro-fuzzy models for early heart attack and cardiovascular risk detection, with particular attention to model architecture, metaheuristic optimization, dataset design, evaluation practice and translation readiness. The review followed the PRISMA 2020 statement and searched Scopus, Web of Science, IEEE Xplore, PubMed, ScienceDirect, SpringerLink and ACM Digital Library for studies published between January 2000 and September 2026. Of 11,321 records identified, 23 studies were retained after deduplication, screening, eligibility assessment and quality appraisal, and their risk of bias was judged with a checklist mapped to the PROBAST domains. Evidence indicates a progression from conventional ANFIS models toward feature-engineered, swarm-optimized, IoMT-enabled and explainability-aware variants. Reported accuracy is often high, but direct ranking is unsafe because Cleveland-derived datasets, which label angiographic coronary artery disease rather than acute myocardial infarction, dominate, external validation remains uncommon, and calibration, subgroup fairness and prospective clinical testing are rarely reported. Optimization methods including particle swarm optimization, genetic algorithms, salp swarm optimization, beetle swarm optimization and hybrid metaheuristics improve search over membership functions, rule parameters and feature subsets, but they also increase computational and reproducibility demands. Rule bases grow exponentially with the number of inputs, and few studies compare ANFIS with tuned ensemble or deep learning baselines under the same validation protocol. The review concludes that future progress should prioritize multi-centre validation, uncertainty quantification, transparent rule extraction, clinically meaningful calibration and reproducible benchmark protocols rather than accuracy alone.