Potato crop disease prediction in Prince Edward Island using machine learning: A decision support approach for farmers

Crop diseases are among the most devastating agricultural challenges that can severely impact productivity and lead to significant crop losses. Spore trapping has long been recognized as a valuable tool in plant pathology for monitoring airborne inoculum; however, its practical use in decision support for farmers remains limited due to cost and region-specific variability. In contrast, existing detection methods often rely on image processing, which primarily detects visible symptoms and therefore lacks the ability to provide early prediction. This study addresses the critical need for disease detection in potato crops by introducing a unique dataset comprising 1482 weekly spore trapping samples collected from 19 agricultural fields in Prince Edward Island (PEI), Canada, spanning from 2019 to 2023. Various Machine Learning (ML) models, including Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), and XGBoost (Extreme Gradient Boost), are employed, in a combination of stratified cross-validation and hyperparameter tuning to enhance predictive performance. Environmental factors such as temperature, dew point temperature, humidity, precipitation, wind speed, and wind direction have been analyzed alongside spore counts. Two potato diseases, Early Blight (EB) and Gray Mold (GM), were predicted for PEI, using disease-specific and combined approach. The group disease model achieved an accuracy of 90% for RF and XGBoost, and 87% for DT. A prototype system was then created to help farmers with early warnings, offering decision-support capabilities for safeguarding crops. This research demonstrates the effectiveness of ML approaches in providing spore-based disease prediction, distinguishing itself from image-based systems by enabling pre-symptomatic forecasts. The findings contribute to real-time disease insights for farmers, helping prevent financial losses, ensuring food security, and promoting sustainable agricultural practices. The future direction of this research includes physical validation as the findings are based on historical spore data.