Abstract
Selecting suitable partners is a critical challenge in Open Innovation because organisations must evaluate multiple technological, relational, strategic, and knowledge-related dimensions while managing uncertainty and information constraints. This study develops a Hybrid Structural Equation Modelling–Machine Learning (SEM–ML) framework integrating theory-guided explanatory modelling with predictive analytics for partner evaluation in innovation networks. The framework is evaluated through three complementary stages: controlled simulation, external domain validation using documented innovation collaborations involving 60 UK firms, and a proxy-based empirical application using World Management Survey (WMS) data from 11,702 organisations. The simulation results show that model performance depends on data structure: Linear Regression performs better under predominantly linear conditions, whereas Random Forest gains predictive advantage when nonlinear and interaction effects are introduced. External domain validation provides real-world grounding for the proposed constructs, with particularly strong documentary evidence for Technological Compatibility and Knowledge Sharing. In the WMS application, Linear Regression slightly outperformed the Hybrid SEM–Random Forest model, while both demonstrated high predictive stability. Knowledge Sharing exhibited the largest predictive contribution to the WMS-derived Partnership Performance proxy, although this is not interpreted as causal influence or theoretical primacy. This study contributes a multi-stage methodological framework that integrates explanatory and predictive analysis while demonstrating that model complexity should be conditional on data structure. The framework provides a foundation for partner-evaluation research, while direct operational validation requires purpose-built inter-organisational partnership data.