Machine Learning-Based Detection of Supply Chain Disinformation: A Systematic Five-Classifier Comparison Using TF-IDF Features — Dataset and Replication Code
Authors/Creators
Description
This repository provides the complete replication package for the manuscript:
"Machine Learning-Based Detection of Supply Chain Disinformation: A Systematic
Five-Classifier Comparison Using TF-IDF Features"
Submitted to Scientific Reports (Nature Portfolio).
CONTENTS:
1. Supplementary_Data_S1_merged_dataset_full.csv — 141 hand-verified labeled
statements (78 TRUE, 63 FAKE) spanning 18 supply chain topic areas (2020–2026).
Sources: DNAINDIA, INDIATODAY, FactCheck.org, PolitiFact, Snopes, BBC Verify,
AFP Fact Check, and government fact-check units.
2. run_all_experiments.py — Complete nested 5-fold cross-validation pipeline for
five classifiers (SVM, LR, Naive Bayes, Random Forest, XGBoost) with TF-IDF
feature extraction. Reproduces all tables and figures in the manuscript.
3. README.md — Full documentation, column definitions, and reproduction instructions.
4. requirements.txt — Python dependencies (Python 3.12, scikit-learn 1.8.0,
XGBoost 3.3.0, scipy 1.17.1).
5. LICENSE.txt — MIT License (code) + CC BY 4.0 (dataset).
KEY RESULT: Linear classifiers (SVM, LR, NB) consistently outperform ensemble
tree methods (RF, XGBoost). LR achieves highest F1 (82.2%); SVM achieves highest
AUC-PR (0.925). Friedman test: χ²(4)=7.435, p=0.115.
Runtime: approximately 8 minutes on a standard CPU.
Files
SupplyChainMgmt_upload_package.zip
Additional details
Dates
- Available
-
2026-08-14