In the urbanization trajectory, the judicious exploitation of underground space has emerged as a pivotal solution to urban challenges. This study addresses the limitations in geological information acquisition and the variability in grading standards within the realm of underground space development. To this end, we introduce an integrated three-dimensional geological suitability evaluation framework that merges machine learning with gray relational analysis. By employing gray relational analysis, we quantify the significance of individual influencing factors, thereby mitigating subjective bias and enhancing the decision-making process. Concurrently, the K-means clustering algorithm is incorporated to increase the precision and impartiality of our evaluation through data point aggregation, yielding clear directives for the suitability assessment of underground space development. In a holistic synthesis, we converge the AHP-entropy weight method, gray relational analysis, and clustering techniques to refine the characterization of the data and the distribution of weights. This integration innovatively optimizes the characterization of data features and the distribution of weights by leveraging the complementary strengths of each method: the AHP-entropy weight method balances subjective expertise with objective data patterns, gray relational analysis handles information incompleteness and defines factor relationships, and clustering provides data-driven, objective classification. This synergistic approach enhances the scientific validity and practical applicability of the evaluation framework. Consequently, our approach provides definitive guidelines for the rational development of underground space, establishing a robust foundation for construction decision-making and propelling subsequent scientific research. The study’s contributions are profound, extending to practical applications and the advancement of academic inquiry within the fields of urban development and geospatial analysis.