<p>The formation of crack kernels, which compromises rice quality, is influenced by the maximum moisture content gradient (MMCG). Consequently, accurately modeling the MMCG as it relates to drying conditions is essential for optimizing drying processes. However, this gradient demonstrates complex, non-linear variations involving multiple variables, making it challenging and time-consuming to model using traditional methods. Artificial neural networks (ANNs) offer a powerful alternative due to their inherent ability to handle such complexities. This work presents an ANN model for predicting the MMCG within rice kernels during combined hot air and far-infrared drying. The model utilized a three-layer, fully connected feedforward network. The inputs were drying time, inlet air temperature, and far-infrared (FIR) intensity. The outputs predicted the average of moisture content (MC), MC at the short axis of the kernel (MCS), and MC at the kernel center, enabling the prediction of MMCG. The two hidden layers, containing 20 neurons, employed a tan-sigmoid transfer function. The Levenberg-Marquardt algorithm was used to train the network. Training data was generated from a finite element method (FEM) simulation based on Fick’s law of diffusion. The trained ANN was validated and tested using randomly generated data. To prevent overfitting, the training process incorporated an early stopping method. The results demonstrate the network’s ability to accurately predict MC and MMCG behavior, as indicated by root mean square error (RMSE) and <i>R</i>-squared (<i>R</i>²). The ANN model demonstrates high predictive accuracy, confirming its effectiveness in modeling moisture content and MMCG during rice drying.</p>