ARTIFICIAL INTELLIGENCE BASED MODELS FOR PREDICTING FOODBORNE PATHOGEN RISK IN PUBLIC HEALTH SYSTEMS

Abstract

This systematic literature review synthesizes evidence on artificial intelligence models used to predict foodborne pathogen risk within public health systems, focusing on how data, methods, and validation practices translate into actionable prevention. We searched major multidisciplinary and domain databases through September 2025 and screened studies against predefined eligibility criteria aligned with PRISMA. A total of 105 peer-reviewed studies met inclusion, spanning outbreak detection, nowcasting and multi-horizon forecasting, spatiotemporal risk mapping, inspection prioritization, and whole-genome sequencing–enabled source attribution. Across the corpus, tree-based ensembles consistently excelled for tabular, establishment-level risk scoring, while recurrent and attention-based sequence models were strongest for delay-aware forecasting. Multi-stream fusion of inspections, laboratory and genomic data, syndromic telemetry, environmental drivers, complaint signals, and supply-chain metadata yielded measurable gains in discrimination, stability, and top-k precision compared with single-stream models. Studies that implemented temporal or geographic external validation, probability calibration, and decision-utility analyses reported smaller but durable improvements that translated into operational benefits such as more critical violations found per fixed inspection budget and earlier detection of emergent clusters at controlled false-alarm rates. Methodological themes associated with credible deployment included leakage-safe temporal and spatial validation, transparent feature engineering, calibrated probabilistic outputs with proper scoring, interpretability to support regulatory scrutiny, drift monitoring with scheduled recalibration, and subgroup assessments to manage equity trade-offs. The review consolidates a decision-oriented framework that maps prediction targets to data and model families, highlights reproducible performance patterns, and outlines governance steps that convert statistical accuracy into sustained public health impact.