The construction industry faces the chronic challenge of project delays, and traditional methods cannot model the complex and nonlinear interactions of factors affecting delays, while machine learning as a data-driven approach can extract hidden patterns from historical data. This study aims to develop a machine learning model for predicting the continuous amount of delay (in days) in Iranian construction projects and identifying the most influential factors; the methodology includes collecting real data from 101 completed projects, implementing seven base algorithms and two ensemble methods, and evaluating performance using MAE, RMSE, and R² metrics. The findings indicate that the Support Vector Regression with linear kernel (SVR) was selected as the final model with an R² of 0.778 and a Mean Absolute Error of 10.4 days. The five variables "material/equipment supply delay", "decision-making speed", "project manager experience", "design revisions", and "payment delay" account for over 73% of the total importance. Three managerial variables comprise more than 42% of the importance, demonstrating that managerial actions have a greater impact on delays than purely technical factors. The proposed model can predict delays of new projects with a Mean Absolute Percentage Error of 12.8%. The final conclusion is that the SVR model, while maintaining simplicity and generalizability, provides practitioners with a practical tool for estimating future delay amounts and focusing on the most effective managerial factors; this data-driven approach can serve as a foundation for developing real-time prediction systems and improving financial discipline in construction projects.