Geng Academic Fraud Detection Report: 'A novel method for corn futures price prediction integrating decomposition, denoising, feature selection and hybrid networks'

Academic fraud report · Geng Detector

Summary

This report assesses Yi Chen's paper published in Annals of Operations Research (2025, DOI: 10.1007/s10479-025-06525-8). The overall verdict is 'highly suspicious.' Three main issues were identified. First, a core logical contradiction: Section 4.1 argues that using today's 'High' price to predict today's 'Close' constitutes information leakage (look-ahead bias), yet Section 5.2's XGBoost feature selection ranks 'High' among the top contributing input features without any lag treatment. Second, the long-term forecasting metrics in Table 11 appear implausible: R² of 0.9402 for a 60-day corn futures forecast is inconsistent with the random-walk nature of commodity time series, and the values form an unusually smooth monotonic decline across horizons. Third, basic terminological errors ('pearman' for Pearson, 'XGBoodt' for XGBoost) suggest minimal editorial review. Findings are limited to textual and tabular inspection; no raw code or data was examined. The verdict indicates high concern but final determination requires investigation by the journal or institution.

Verdict

🟠 Highly suspicious. Multiple internal contradictions and implausible empirical results suggest potential data leakage, fabricated figures, or hasty assembly. Confirmation requires raw data and code from the authors.

Key findings

Logical contradiction on 'High' feature usage: The paper explicitly warns against using today's 'High' price to predict today's 'Close' (information leakage), yet lists High as a selected input feature with non-zero XGBoost importance in Section 5.2.
Implausibly perfect long-term forecasts: Table 11 reports R² = 0.9402 for 60-day-ahead corn futures prediction and R² = 0.8852 for the CBA baseline, with an unusually smooth decline from R² = 0.9871 (5-day) to 0.9402 (60-day). Such performance contradicts established financial time-series theory regarding the random-walk behavior of commodity prices.
Basic terminology and spelling errors: 'pearman correlation coefficient' (instead of Pearson) and 'XGBoodt' (instead of XGBoost) appear in the published version, indicating insufficient proofreading.

Evidence highlights

Section 4.1 quote (paraphrased): using today's High price to predict today's Close is invalid because the High is unknown before close.
Section 5.2 XGBoost selection output includes High among 11 non-zero importance features (Low, Open, CSZL.DCE, MA, Spot price, High, BYQport, ...).
Table 11 long-term forecast metrics:
5-day: R² = 0.9871
10-day: R² not specified in report
60-day: CBA-X-PVW R² = 0.9402, RMSE = 23.05
60-day CBA baseline R² = 0.8852
DOI: 10.1007/s10479-025-06525-8
Journal: Annals of Operations Research (2025) 353:449–484, published online 18 February 2025.
Author affiliation: Yi Chen, Sun Yat-sen University Business School.

Notes

This report is AI-assisted and intended for academic discussion only. Final determination of misconduct requires formal investigation.
Only the published manuscript was examined; no raw training data or source code was accessed.
Alternative benign explanations (e.g., the authors intended a lagged High without stating so) cannot be ruled out without supporting documentation from the authors.