Prelecionista: Federico José González Villasanti
Orientador: Eduardo Seiti Gomide Mizubuti
Data: 29/09/2023, às 14h, online*

 Resumo:  Tomato Late Blight (TLB) caused by Phytophthora infestans (Mont.) de Bary is one of the most destructive diseases of tomato crops (Solanum lycopersicum). Due to its economic importance, several integrated management tools were developed to improve its control, including disease forecasting models. Available models in the market rely mostly on weather-based risk alerts and empirical approaches, while recent technologies such as machine learning provide new capabilities for modeling and forecasting. Six field trials in two years were conducted to gather disease measurements. Each trial had a hyperlocal weather station installed to gather meteorological data. A Support Vector Machine (SVM) model was used to forecast disease onset with an accuracy of 95%. Two machine learning models constructed to forecast TLB progress were tested and compared: Random Forest Regressor (RF) and an Extreme Gradient Boosting Regressor (XGBR). The XGBR returned a lower symmetric Mean Absolute Percentage Error when compared to the RF for the exponential stage of the epidemics and a similar error for the asymptote stage. The weather variables that affected TLB progress were related to water availability. ML models can predict the onset and development of TLB, despite clear limitations regarding a small disease dataset. Machine learning models can be used to forecast disease and become part of a disease support system aimed at improving TLB management.

*Público externo ao PPG interessado em receber o link, favor enviar email com sua identificação para federico.villasanti@ufv.br.