Telecom churn – reproducibility package (code, calibrated scores, models, figures)

This record provides code and artifacts to reproduce the manuscript’s results. It includes source code (training/evaluation, post-hoc XAI), calibrated out-of-fold/test scores, cross-validation summaries, trained models, and all figures. No resampling is applied (class imbalance handled via class_weight="balanced"); PR-AUC is the primary selection metric; probabilities are isotonic-calibrated for decision-grade risk tiering. Raw IBM Telco Customer Churn data are not redistributed; obtain from the IBM Sample Data Repository.本数据集包含用于复现投稿手稿研究结果的代码与配套资源，具体涵盖：源代码（训练/评估、事后可解释AI（post-hoc XAI））、经校准的折外/测试集评分、交叉验证结果汇总、训练完成的模型以及全部可视化图表。本数据集未执行重采样操作（通过将class_weight设为"balanced"处理类别不平衡问题）；精确召回曲线下面积（PR-AUC）为首要选择指标；概率经保序校准（isotonic-calibrated）后用于决策级风险分层。原始IBM电信客户流失（IBM Telco Customer Churn）数据集不予重新分发，请从IBM示例数据存储库获取该原始数据。