When Does Cost-Sensitive Learning Help in Predictive Maintenance? A Cross-Dataset Benchmark Study — Code and Data

Complete experimental code, per-fold raw scores, and aggregated result tables for the paper When Does Cost-Sensitive Learning Help in Predictive Maintenance? A Cross-Dataset Benchmark Study (Neural Computing and Applications, under review, 2026). The study benchmarks nine learners under one identical protocol (10-fold cross-validation repeated three times) across three predictive-maintenance datasets spanning a difficulty gradient — MetroPT-3 (separable), AI4I-2020 (moderate), and SCANIA Component X (hard) — plus CWRU as a saturated cautionary case, using a Normalised Cost Savings (NCS) metric comparable across datasets. The archive reproduces every table and figure in the paper.本资源收录论文《成本敏感学习何时助力预测性维护？一项跨数据集基准研究》（*Neural Computing and Applications*，2026年待刊）的完整实验代码、各折次原始得分与汇总结果表格。 本研究采用统一实验协议（10折交叉验证重复三次）对9种学习器开展基准测试，测试涵盖3个按难度梯度排布的预测性维护数据集：MetroPT-3（可分离型）、AI4I-2020（中等难度）与SCANIA Component X（高难度），并额外纳入CWRU作为饱和警示案例，同时使用可跨数据集横向比较的归一化成本节约值（Normalised Cost Savings, NCS）作为评价指标。本归档文件完整复现了论文中的全部表格与图表。