Abstract: In the field of comprehensive evaluation of construction projects, the issues of one-sided quantification of indicators and subjective evaluation are prominent, and the influence of human judgment is difficult to ignore. To reduce the impact of human judgment on evaluation work and ensure efficient, accurate, and objective comprehensive evaluation results, this study, based on the multi-level fuzzy algorithm and RBF neural network model, combines multi-faceted indicators and data of construction projects to construct a multi-indicator construction project evaluation model, thereby facilitating evaluation tasks such as enterprise scoring, project comparison and selection, and risk assessment. The results show that the evaluation model can effectively carry out quarterly analysis: verified by 17,774 quarterly objective evaluations, it has significantly increased the number of high-quality enterprises, providing data support for enterprise selection, scheme review, and comparison during bidding. The evaluation model can also achieve risk assessment: after training on 1008 indicators from 18 types of projects and 477 risk samples, the accuracy rate of risk level determination exceeds 80%, with no cross-level misjudgment, effectively compensating for potential determination errors caused by insufficient professional knowledge. It comprehensively assists managers in strengthening risk control and improving project inspection efficiency. The comprehensive evaluation model of construction projects based on the multi-level fuzzy algorithm and RBF neural network has been successfully applied in the construction process of Guangzhou's construction projects, providing a unified and objective evaluation standard for joint review and decision-making selection of project schemes, enterprise evaluation grading, and risk level classification of engineering hidden dangers. This effectively avoids subjective influences and errors caused by human assessment.