Flood susceptibility mapping is one of the most effective approaches for flood risk management and reducing damages caused by flood events. The present study aimed to evaluate and compare the performance of the CNN-sim and GMDH models integrated with the Fuzzy-Weight of Evidence (Fuzzy-WofE) approach for flood susceptibility mapping in the Gorganrud watershed. Fourteen environmental factors, including elevation, rainfall, slope, lithology, land use, soil texture, drainage density, distance from rivers, distance from roads, curvature, Topographic Wetness Index (TWI), Stream Power Index (SPI), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Moisture Index (NDMI), were used as conditioning variables. The Weight of Evidence (WofE) method was employed to determine the initial weights of the factors, and the resulting values were normalized using the Min–Max fuzzy membership function. The weighted variables were then used as inputs for the CNN-sim and GMDH models. The results indicated that elevation, rainfall, and SPI were the most influential factors in flood occurrence, with fuzzy weights of 1.000, 0.8793, and 0.8546, respectively, while soil texture had no significant contribution to flood susceptibility prediction. Model evaluation demonstrated that the CNN-sim model outperformed the GMDH model across all performance metrics. The CNN-sim model achieved an AUC of 0.9274, accuracy of 0.8448, precision of 0.8387, recall of 0.8667, and F1-score of 0.8525, whereas the GMDH model yielded an AUC of 0.8476 and accuracy of 0.6207. Analysis of the flood susceptibility map generated by CNN-sim showed that 51.66% of historical flood locations were situated within the very high susceptibility class and 21.67% within the high susceptibility class, confirming the strong predictive capability of the model.Overall, the findings demonstrate that integrating the Fuzzy-WofE approach with deep learning techniques significantly enhances flood susceptibility prediction accuracy.