This study presents a robust application of Artificial Neural Networks (ANNs) integrated with four swarm intelligence algorithms—DE, FF, EO, and BBO for predicting the swell pressure (Ps) of expansive soils. Each ANN-based model was trained using nine key geotechnical parameters sourced from published literature. Model performance was assessed using multiple statistical metrics, including MAE, RMSE, R2, NS, MAPE, IA, VAF, PI, and IOS. All models demonstrated strong predictive capability with R2 values exceeding 0.8; however, the ANN-FF model outperformed others, achieving the highest R2 values and lowest error values for both training and testing phases. Residual error analysis confirmed the superior accuracy and consistency of the ANN-FF model in predicting swell pressure, with ANN-BBO performing reliably, while ANN-DE and ANN-EO showed reduced precision and higher variability. These findings highlight the ANN-FF model's effectiveness in optimizing ANN hyperparameters and its potential for practical application in real life application.