Numerous Iranian construction projects are left unfinished for various reasons, one of the most significant being incorrect cost estimation, particularly during the initial design phase. Cost prediction is among the most critical stages of civil project planning and a necessary objective for project managers, as deviation from initial estimates leads to numerous issues and legal disputes. Furthermore, controlling costs while maintaining the specified quality is the single most critical factor for continuing civil project activities, considering the heightening competition among contractors and diminishing profit margins. Thus, cost prediction is a crucial tool for controlling cost deviations for all project stakeholders, ranging from the owners to the contractors. This study aims to extract main criteria influencing cost prediction using the Delphi method, and predict construction cost using artificial neural network for residential building of Mashhad city based on 70 projects data. The findings indicate that the best model trained using neural networks has a correlation coefficient of 0.87, demonstrating the satisfactory performance of the model. Additionally, principal component analysis demonstrates that parameters such as usable area per floor, ground floor area, built-up area, the total number of units, concrete volume used in the building, construction duration, exterior wall area, and land area are first-ranked factors, while the number of floors above ground level, total floors from the foundation, and building height from the foundation are second-ranked factors. Sensitivity analysis based on obtained results revealed that exterior wall area and the number of above-ground level floors were the most significant factors in this paper. Using this method can provide employers and contractors of construction projects with a quick and appropriate estimation for the initial cost of constructing residential projects.