Bankruptcy Prediction Analysis of Companies In Indonesia Using Ensemble Learning Methods to Improve Prediction Accuracy

Corporate bankruptcy poses significant economic risks, particularly amid post-pandemic uncertainties. Traditional models often fail to capture the nonlinear complexity of financial data. This study improves bankruptcy prediction accuracy for Indonesian companies using ensemble learning techniques, comparing Random Forest, Logistic Regression, and XGBoost. A Systematic Literature Review (SLR) of 23 peer-reviewed studies is combined with financial data from 80 publicly listed Indonesian companies (40 bankrupt, 40 non-bankrupt) covering the period from 2015 to the first quarter of 2025. Models are evaluated using accuracy as the primary metric, with precision, recall, and F1-score as supporting metrics to provide a more comprehensive performance assessment, while feature importance analysis highlights key liquidity and profitability indicators, such as the current ratio, quick ratio, and return on assets (ROA). Results show that XGBoost and Random Forest achieved 93% accuracy, while Logistic Regression achieved 86.7% accuracy, confirming that ensemble models consistently outperform other models, whereas Logistic Regression remains interpretable. Findings provide practical implications for early warning systems, risk assessment, regulatory monitoring, and governance, offering actionable guidance for stakeholders and contributing to the development of data-driven bankruptcy prediction frameworks.企业破产（Corporate bankruptcy）会带来显著的经济风险，在后疫情时代的不确定性背景下尤为突出。传统模型往往难以捕捉金融数据的非线性复杂特征。本研究针对印尼企业，通过集成学习（ensemble learning）技术提升破产预测准确率，对比了随机森林（Random Forest）、逻辑回归（Logistic Regression）与XGBoost三种模型。本研究整合了针对23篇同行评议研究的系统文献综述（Systematic Literature Review, SLR），以及2015年至2025年第一季度期间80家印尼上市公司的财务数据——其中破产与非破产企业各40家。模型以准确率作为核心评估指标，辅以精确率、召回率与F1值以实现更全面的性能评估；同时通过特征重要性分析，明确了流动性与盈利性相关的关键指标，例如流动比率、速动比率与资产收益率（Return on Assets, ROA）。实验结果显示，XGBoost与随机森林的准确率达到93%，逻辑回归的准确率为86.7%，证实集成模型的整体性能优于其他模型，而逻辑回归仍具备可解释性。本研究结果可为早期预警系统、风险评估、监管监测与公司治理提供实践参考，为相关利益相关者提供可落地的指导，并助力数据驱动型破产预测框架的发展。