<p>Housing price forecasting is a critical aspect of societal functioning as it has a profound impact on individuals, societal stability, and government management. Accurate predictions enable informed decision-making during house purchases. This study introduces a novel hybrid model called BoostingCNNSOS, which combines the strengths of convolutional neural network (CNN) and boosting algorithms to deliver highly accurate housing price forecasts with minimal errors. To further enhance the model’s performance, an optimized meta-heuristic method is employed for precise parameter adjustments. The analysis utilizes a comprehensive dataset of 998,299 sale transactions of residential units in Tehran from 2014 to 2021. This dataset includes 11 crucial features that are essential for accurate forecasting. Before the analysis, a series of preprocessing steps are undertaken to ensure the integrity and reliability of the data. The BoostingCNNSOS model’s effectiveness is evaluated by comparing it with established models such as XGBoost, CatBoost, LightGBM, and LSTM in terms of accuracy and error rates. Various evaluation criteria, including MSE, RMSE, MAE, ME, and <i>R</i>², are employed. The consistent comparisons across all evaluation criteria demonstrate the stability and superior performance of the BoostingCNNSOS model. In Fold-1, the model achieves exceptional performance, with MSE, RMSE, MAE, ME, and <i>R</i>² values of 0.0000025, 0.00158, 0.00046, 0.1364, and 0.88, respectively. These results highlight the accuracy and predictive capabilities of the model in housing price forecasting. The BoostingCNNSOS model continues to demonstrate superiority in Fold-2 to Fold-4. In contrast, the LSTM model shows weaker performance, while the XGBoost model showcases relatively good performance compared to other models. The findings of this research hold significant value for various stakeholders, including builders, planners, policymakers, researchers, and home buyers.</p>