Abstract: Insufficient safety investment on construction projects hinders their safety performance. The present study therefore proposes a theoretical framework of "Institutional Isomorphism, Safety Climate and Safety Investment" by integrating the Institutional Isomorphism Theory and the Safety Climate Theory. We employed correlation analysis and the random forest machine learning model on data from 953 valid questionnaires. Results demonstrate that safety climate serves as the core determinant of safety investment behavior, with organizational and team safety climate jointly accounting for over 86.00 % of the explanatory power. In contrast, institutional isomorphism exerts weak direct effect on safety investment. However, it poses significant mediating effect via safety climate. Its mediating pathway contributing 21.80 % to the total impact, with normative isomorphism and mimetic isomorphism demonstrating notable effects. The present random forest model achieves an R² value of 0.814, outperforming conventional linear analysis approaches in unraveling the complex nonlinear coupling relationships across multiple variables. This study probes into the application of machine learning methods in construction safety management and delineates the transmission mechanism underlying "institutional pressure, cultural response and investment behavior". It provides theoretical and practical implications for construction enterprises to synergistically optimize institutional compliance and cultivate safety cultures, enhancing the efficiency of safety resource allocation.