AI Based Prediction of Crop Loss Risk in Tropical Agricultural Systems of Panama with Implications for Food Security Planning

This study evaluated a Random Forest model for estimating crop-loss risk in small- and medium-scale maize and rice production systems in Los Santos, Guararé, and Tonosí, in the Azuero region of Panama. The dataset comprised approximately 720 repeated monthly records from 20 production units across three agricultural cycles, combining climatic, agronomic, phytosanitary, and production information. Crop loss was expressed relative to a common yield reference of 6.0 t/ha, with operational categories of low (<10%), medium (10–25%), and high (>25%) risk. The analysis used an 80/20 training-validation partition and five-fold cross-validation. Internal validation yielded R2 = 0.78, while the fold-specific R2 values shown in the cross-validation output ranged from approximately 0.75 to 0.80. Accumulated precipitation (28.6%) and average temperature (22.1%) had the greatest model importance, followed by irrigation frequency (15.4%) and pest incidence (13.7%). A matrix-based classification analysis of 81 observations yielded 64 correct risk assignments (79.0%), with an 85.7% high-risk recall. These findings support localized crop-loss risk assessment within the participating production units and identify potential applications in irrigation management, pest surveillance, and planting decisions relevant to food-security planning.