Machine learning models for plant fungal disease detection and risk prediction: A systematic review

Plant fungal diseases pose significant threats to global agriculture, undermining food security and economic stability. Traditional detection methods often fail to identify infections during latent stages, highlighting the need for proactive solutions. This systematic review synthesizes advancements in machine learning (ML) for fungal disease detection, classification, and forecasting, analyzing 22 studies (2010–2025) through PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Results reveal that ML models, particularly image-based convolutional neural networks (CNN) and sensor-fused Random Forest (RF) algorithms, achieve high accuracy in disease classification and detection (>95% in postharvest applications) and show promise in pre-symptomatic detection and risk forecasting (67–93% accuracy). Hybrid models integrating environmental, spectral, and genomic data demonstrated promise in capturing complex host-pathogen interactions, while transfer learning reduces dependency on large datasets. However, challenges persist, including regional data biases, computational costs, infrastructural barriers, and limited model interpretability. Geographically, studies disproportionately focus on temperate regions and staple crops, neglecting perennial and specialty plants. Future efforts must prioritize globally representative datasets, lightweight ML architectures for resource-limited settings, and explainable AI to foster farmer trust. If the identified challenges particularly limited field validation and infrastructural barriers are systematically addressed, ML holds significant potential to transition from experimental innovation toward a foundational element of sustainable, precision agriculture, potentially enabling more timely interventions and reducing reliance on broad-spectrum agrochemicals. This review uniquely evaluates current limitations and advocates for interdisciplinary collaboration to ensure equitable, scalable deployment in diverse agroecosystems, underscoring ML’s transformative potential for sustainable agriculture.