A Multi-Stage Modeling Framework Integrating Random Forests and Cluster Analysis: Feature Identification and Prediction for the Global Innovation Index

Abstract

Based on data from the Global Innovation Index, this paper constructs an integrated data-driven framework that combines feature selection, cluster analysis, and predictive modeling. First, a random forest model is used to assess the importance of multidimensional indicators, identifying key variables from high-dimensional features to effectively capture complex nonlinear relationships. Second, the K-means method is employed to stratify the innovation levels of different countries, achieving structured classification. Building on this foundation, a sliding-window-based random forest regression model is introduced to transform short-sequence data into supervised learning samples, enabling high-precision forecasting of future indicators. The model demonstrates strong fitting capability and stability in testing. The core innovation of this study lies in the systematic integration of feature selection, structural classification, and predictive modeling into a unified analytical workflow, thereby enhancing the model’s adaptability to complex data and its interpretability. This method avoids reliance on a single model and demonstrates good generalization performance under various data structural conditions, making it applicable to scenarios involving multi-indicator comprehensive evaluation and dynamic forecasting. Overall, the proposed model framework enhances forecasting accuracy while simultaneously providing a synergistic characterization of key drivers and structural features, offering a universal and efficient technical approach for the analysis of complex systems.