Predicting costs in modular construction: A comprehensive study using SSA and SVR combined models

Prefabricated construction offers a sustainable alternative to traditional methods by reducing environmental pollution and resource waste. However, its adoption is limited by high upfront costs, estimation inaccuracies, and scarce engineering data. This study proposes an accurate cost prediction approach combining Random Forest (RF) for feature selection and Support Vector Regression (SVR) optimized by the Sparrow Search Algorithm (SSA).Model performance is evaluated using MAE, MSE, and RMSE under a 10-fold cross-validation framework. Results show that RF-based feature selection improved prediction accuracy, while SSA outperformed GA and GWO in hyperparameter optimization. Compared with the baseline SVR model, the proposed RF-SSA-SVR model achieves lower average prediction errors under cross-validation. In addition, representative case-level prediction results show that the relative errors remain at a low level, satisfying the ±20% accuracy requirement for investment estimation at the feasibility study stage of prefabr icated residential projects.This method provides a reliable framework for early-stage cost prediction and supports investment decision-making in prefabricated construction.