Research on Financial Risk Early Warning Based on Stacked Random Forest Algorithm

Financial risk early warning is a critical area in the banking and corporate sectors, aimed at predicting potential defaults or financial instability before they occur. Traditional machine learning methods, while effective, often face challenges such as overfitting, insufficient generalization across different industries, and limited adaptability to complex, highdimensional financial data. To address these issues, this research proposes a stacked Random Forest algorithm that integrates multiple Random Forest classifiers with varied hyperparameters and training subsets, combined through a meta-classifier to enhance prediction accuracy and robustness. The approach leverages diverse learning patterns from multiple base learners, reducing bias and variance while improving the model's sensitivity to early risk indicators. The study employs publicly available datasets, including the Companies Financial Risk Dataset and the UCI German Credit Dataset, which contain comprehensive financial indicators such as liquidity ratios, profitability metrics, debt levels, and credit history attributes. The stacked RFC framework first trains several base-level Random Forest models on different feature subsets or bootstrap samples, and then feeds their probabilistic outputs into a meta-level Random Forest, which produces the final early warning signal. Experimental results demonstrate that the proposed model achieving higher accuracy of 97.2 %, particularly in imbalanced financial risk scenarios. The stacked RFC method shows strong generalization ability, offering a reliable and scalable tool for early financial risk detection in real-world applications.