["Transactions in GIS, Volume 30, Issue 4, June 2026. ", "\nABSTRACT\nWildfire susceptibility mapping (WSM) is critical for forest management, land‐use planning, and disaster risk mitigation. Although hybrid artificial neural network (ANN) models optimized by metaheuristic algorithms are increasingly used in susceptibility mapping, they are often evaluated without strong machine learning benchmarks, spatially robust validation, or statistical significance testing. This study benchmarks two practically parameter‐free evolutionary ANN models, Bernstein–Levy Differential Evolution ANN (BDE‐ANN) and Multi‐Population Differential Evolution ANN (MDE‐ANN), against Logistic Regression, Support Vector Machine with RBF kernel, Random Forest, XGBoost, ANN, and standard DE‐ANN. A balanced spatial dataset consisting of 272 wildfire and 272 non‐wildfire locations in Çanakkale, Türkiye, was constructed using 14 geo‐environmental conditioning factors. After multicollinearity assessment and ablation analysis, average temperature was excluded, resulting in a final set of 13 predictors. To reduce optimistic bias caused by spatial autocorrelation, model performance was evaluated using spatial block cross‐validation. Predictive uncertainty and pairwise model differences were further assessed through bootstrap AUC confidence intervals, Wilcoxon signed‐rank tests, and McNemar tests. The results showed that MDE‐ANN achieved the highest overall discrimination and lowest prediction error (Test AUC = 0.874 ± 0.064; MSE = 0.120 ± 0.041). However, its advantage over Random Forest and XGBoost was not statistically significant, indicating that MDE‐ANN should be interpreted as a top‐tier but not universally dominant classifier. In contrast, BDE‐ANN provided the highest recall (0.985 ± 0.030) and F1‐score (0.842 ± 0.052), making it particularly suitable for recall‐priority screening where missed fire‐prone areas are highly undesirable. RF and XGBoost offered highly competitive performance with substantially lower computational cost. Overall, the findings support a task‐oriented WSM framework in which MDE‐ANN is preferable for balanced risk discrimination, BDE‐ANN for fire‐detection‐oriented screening, and tree‐based ensembles for rapid baseline deployment.\n"]