Explainable AI for Predicting Latent Period and Infection Stage Progression in Tomato Fungal Diseases

Accurate prediction of the latent period and disease progression in tomato fungal infections is critical for enabling timely interventions and effective disease management. Unlike existing AI-based approaches that primarily classify diseases after symptom emergence, this study innovates by predicting infection stages from the asymptomatic (latent) phase through complete symptom development, integrating biologically grounded feature extraction with explainable artificial intelligence (XAI). This study presents a novel, XAI framework capable of day-wise prediction of infection stages, including the latent period, for four major fungal pathogens in tomatoes: Alternaria alternata, Alternaria solani, Botrytis cinerea, and Fusarium oxysporum. A high-resolution (Red-Green-Blue) RGB image dataset was collected under controlled inoculation conditions, capturing daily changes in infected and healthy tomato leaves over six days post-infection. The pipeline included image preprocessing, lesion segmentation, and extraction of biologically meaningful features (texture, color, and shape) reflecting underlying physiological changes in the plant. Feature relevance across infection stages was dynamically assessed using the Relief algorithm, providing interpretability by linking visual changes to disease biology. Machine learning classifiers, Support Vector Machine (SVM) and Random Forest (RF), were optimized using Particle Swarm Optimization (PSO), achieving significant improvements in infection day prediction accuracy across all four pathogens. For example, RF accuracy increased from 76.14% to 94.17% for A. alternata (with 97.96% sensitivity and 99.48% specificity on day 6 post-inoculation) and from 80.01% to 97.08% for B. cinerea. Critically, the model accurately identified the latent period for each pathogen, detecting microscopic texture changes on day 1 post-inoculation when no visible symptoms were present. By bridging the gap between AI and plant pathology, this framework enables early diagnosis of fungal diseases with explainable outputs. The approach offers a scalable, non-destructive, and biologically grounded tool for integrated disease management, with potential applications across diverse crops in precision agriculture.