<p>Proactively managing supplier performance in the volatile Integrated Circuit (IC) design industry is critical for supply chain resilience. However, traditional evaluation methods are retrospective and often fail to handle the inherent data uncertainty and complex operational structures. To address this gap, this study develops and empirically validates a novel predictive analytics framework that integrates Chance-Constrained Network Data Envelopment Analysis (CC-NDEA) with machine learning. First, we propose a two-stage CC-NDEA model that uniquely captures the interplay between operational management efficiency (cross-firm average across all eight years: 0.808) and R&amp;D efficiency (cross-firm average across all eight years: 0.731) processes while accommodating data uncertainty and negative values (e.g., profit). This model is applied to an eight-year panel dataset of 56 Taiwanese IC design firms. The resulting robust efficiency scores serve as theoretically-grounded features for four machine learning algorithms—LSTM, GRU, Random Forest, and Gradient Boosting—to forecast future performance. Our empirical results reveal that Gradient Boosting excels at predicting technical and R&amp;D efficiencies (R²&gt; 0.73), while LSTM is superior for operational efficiency. This study advances the literature by developing an integrated solution that holistically combines uncertainty, network structures, negative data handling, and machine learning for supplier evaluation. While prior studies have addressed individual dimensions or pairs thereof, our work is among the first to unify all four within a predictive supplier evaluation context. Ultimately, our framework transforms static efficiency assessment into a dynamic, forward-looking tool, enabling managers to make proactive decisions to optimize the supply chain.</p>