Explainable adaptive ensemble learning with imbalance mitigation for manufacturing sector financial risk warning

Financial risk early warning is crucial for safeguarding manufacturing enterprises’ stability, yet existing methods face significant challenges in handling the inherent nonlinear dynamics and chaotic fluctuations within imbalanced financial data while maintaining interpretability. Tree ensemble approaches predominantly rely on static resampling techniques that distort the natural fractal structures of financial distributions and fail to capture emergent patterns of rare distress signals in nonlinear financial phase spaces. Moreover, conventional ensemble-based systems suffer from opaque decision-making processes that obscure underlying nonlinear dynamics, impeding extraction of actionable insights from complex bifurcation patterns in corporate financial trajectories. To address these nonlinear system challenges, we propose an interpretable adaptive tree ensemble (IATE) for interpretable manufacturing financial systems. The approach incorporates a probability-guided adaptive under-sampling algorithm that preserves essential statistical properties of the nonlinear financial manifold while dynamically rebalancing class distributions to capture fractal geometry of distress patterns. A heterogeneity-aware ensemble mechanism synergizes multi-perspective financial features through posterior probability aggregation in nonlinear feature space, enhancing robustness against data imbalance. Furthermore, a dual-interpretability system combines global feature importance metrics with financial indicator-optimized partial dependence plots that explicitly reveal nonlinear risk drivers and their complex interactions within the financial phase space. Empirical evaluation on manufacturing firm-year observations demonstrates IATE’s superior performance in detecting chaotic signatures of financial distress, achieving significant improvement in distress signal detection across 4 predictive periods compared to XGBoost while maintaining robust discriminative capability amid nonlinear financial fluctuations. The interpretability components successfully identify actionable precursors within the complex financial dynamical system, enabling proactive interventions that account for nonlinear evolution of manufacturing enterprises’ financial risk trajectories.