Supply chain decision-making faces growing vulnerabilities from disinformation, where false or misleading statements can trigger demand forecasting errors, inventory mismanagement, and significant procurement disruptions. To address these operational hazards, researchers have conducted a systematic evaluation of five machine learning classifiers designed for automated supply chain disinformation detection. The investigation specifically benchmarks the capabilities of support vector machines, logistic regression, Naive Bayes, random forest, and XGBoost algorithms.
To establish an empirical foundation for model training and evaluation, the research team constructed a hand-verified dataset consisting of 141 labeled statements obtained from professional fact-checking organizations. These evaluated statements span 12 distinct supply chain topic areas to capture diverse forms of operational misinformation. The methodological framework employed nested stratified 5-fold cross-validation to ensure rigorous validation standards across all tests. Textual data preprocessing and feature representation were conducted using term frequency-inverse document frequency feature extraction, allowing the five machine learning models to classify textual statements based on statistical term distributions.
Experimental evaluations revealed that linear classifiers substantially outperform complex ensemble tree methods across multiple performance metrics. Logistic regression achieved the highest overall F1-score at 82.2 percent and demonstrated the greatest precision at 89.0 percent among all evaluated models. Meanwhile, support vector machines attained the highest area under the precision-recall curve at 0.925 as well as the highest receiver operating characteristic area under the curve at 0.918. Naive Bayes recorded the highest recall at 84.2 percent, confirming strong sensitivity for detecting misleading statements. In contrast, ensemble tree architectures lagged behind linear counterparts, with random forest and XGBoost recording F1-scores of only 72.4 percent and 72.5 percent, respectively.
Statistical assessment using a Friedman test yielded a test statistic of chi-square(4) = 7.435 with a p-value of 0.115, indicating no statistically significant difference among the five classifiers at the current dataset sample size. Nevertheless, calculated Cohen’s d effect sizes reached values up to 2.16, demonstrating a practically meaningful performance advantage for linear models over ensemble tree techniques. To address potential bias, the authors conducted a sensitivity analysis on a 104-row subset limited strictly to supply chain statements, confirming that the comparative performance patterns remain robust against source-domain confounding.
Based on these empirical findings, the researchers outline specific deployment recommendations depending on organizational objectives in supply chain operations. Support vector machines are specifically recommended for operational environments where threshold flexibility is required to adjust classification boundaries dynamically. Conversely, logistic regression is recommended when minimizing false positives represents the primary operational priority for supply chain decision-makers. Together, these results demonstrate that streamlined linear classification models provide practical and robust tools for defending supply chain forecasting, inventory management, and procurement planning against misleading information.
Subject of Research: Automated detection of disinformation in supply chain decision-making using machine learning classifiers
Article Title: Disinformation-aware decision-making in supply chains using support vector machines
Article References: Dolly, B., Jindal, S. K., Kumar, N., Kumar, V., Dwivedi, S., & Nandal, A. (2026). Disinformation-aware decision-making in supply chains using support vector machines. Scientific Reports. https://doi.org/10.1038/s41598-026-68744-8
Image Credits: AI Generated
DOI: 10.1038/s41598-026-68744-8
Keywords: supply chain management, disinformation detection, support vector machines, logistic regression, naive bayes, random forest, xgboost, linear classifiers, ensemble methods, tf-idf feature extraction, nested cross-validation, precision and recall, f1-score, roc-auc, auc-pr, demand forecasting, inventory management, procurement disruptions
Cite Scienmag News
Teresa Odom. (November 18, 2026). Linear Classifiers Outperform Tree Models in Detecting Supply Chain Disinformation. Scienmag. https://scienmag.com/linear-classifiers-outperform-tree-models-in-detecting-supply-chain-disinformation/
Teresa Odom. "Linear Classifiers Outperform Tree Models in Detecting Supply Chain Disinformation." Scienmag, 18 November 2026, https://scienmag.com/linear-classifiers-outperform-tree-models-in-detecting-supply-chain-disinformation/. Accessed 18 November 2026.
Teresa Odom. "Linear Classifiers Outperform Tree Models in Detecting Supply Chain Disinformation." Scienmag. November 18, 2026. https://scienmag.com/linear-classifiers-outperform-tree-models-in-detecting-supply-chain-disinformation/








