DBFC-1: Database Forensic Classification Corpus and DBF-Net Reference Model (v1.0)

A fully reproducible benchmark corpus of 24,000 reconstructed database events for AI-enabled database forensic triage, spanning one benign class and five tamper categories (tamper, exfiltration, anti-forensic, privilege-escalation, destructive), with an 18-feature forensic taxonomy across seven groups, a deterministic generator (seed 20260115), official train/validation/test splits, the DBF-Net gradient-boosted reference classifier, three baselines, full evaluation code (significance tests, ablation, evasion, SHAP), and all figures. Accompanies the paper "From Artifacts to Attribution: An AI-Enabled Framework for Database Forensics with the DBFC-1 Public Benchmark Corpus." Data and figures are licensed CC BY 4.0; code is licensed MIT.本基准语料库具备完全可复现性，包含24000条经重构的数据库事件，面向人工智能（AI）赋能的数据库取证甄别任务；涵盖1个良性类别与5类篡改场景：篡改（tamper）、数据外泄（exfiltration）、反取证（anti-forensic）、权限提升（privilege-escalation）以及破坏性操作（destructive）。该语料库构建了覆盖7个组别、包含18项特征的取证分类体系，附带基于固定种子（20260115）生成的确定性生成器，提供官方预设的训练集、验证集与测试集划分方案，配备DBF-Net梯度提升参考分类器，包含3个基准模型、完整的评估代码库（涵盖显著性检验、消融实验、规避测试与SHAP可解释性分析）以及全部配套图表。本语料库配套发表于论文《从痕迹到归因：面向数据库取证的人工智能赋能框架与DBFC-1公开基准语料库》（"From Artifacts to Attribution: An AI-Enabled Framework for Database Forensics with the DBFC-1 Public Benchmark Corpus"）。数据与图表采用知识共享署名4.0（CC BY 4.0）许可协议，代码采用MIT许可协议。