A MATHEMATICAL MODEL FOR MULTI-CLASS PREDICTION OF AN INTELLIGENT SWOLLEN SHOOT DISEASE SYSTEM BASED ON AGRO-ENVIRONMENTAL DATA

The cocoa sector, a major economic pillar in West Africa, is facing a concerning phytosanitary threat: Cocoa Swollen Shoot Virus (CSSV), mainly transmitted by mealybugs. This viral disease leads to a progressive decline of plantations, affecting yields and compromising the socio-economic resilience of producers. Faced with the limitations of traditional detection methods, which are often delayed, this study proposes an innovative AI-based approach for predicting CSSV infection stages. Using a multinomial logistic regression model, 250 cocoa plants were simulated with eight agro-environmental variables including plant age, proximity to infection source, soil moisture, ambient temperature, planting density, varietal resistance, phytosanitary treatment, and mealybug level. The supervised classification achieved a recall of 77.4% for healthy plants, but weaker performance for infected classes (21.4% and 6.3%), reflecting the complexity of intermediate stages. An interactive map was generated to geographically visualize predictions, facilitating risk zone prioritization. Additionally, 30-day dynamic simulations showed over 80% loss of healthy plants without intervention. These results confirm the relevance of AI as a predictive tool capable of guiding farmers in their health decisions. Future perspectives include the integration of multi-source data, training on real data, and deployment of embedded tools for predictive, optimized, and sustainable agricultural e-health.