Academic Fraud Investigation Report: Chen (2025), 'A novel method for corn futures price prediction integrating decomposition, denoising, feature selection and hybrid networks'

Academic fraud report · Geng Detector

Summary

Verdict: Severe methodological and data integrity concerns (red flag). The paper proposes a hybrid model (CBA-X-PVW) for corn futures price forecasting and reports that multi-step (5-day, 10-day, 20-day, 40-day, 60-day) predictions outperform next-day (1-day) predictions, which violates the basic statistical property that forecast error should increase with horizon. The author's proposed model shows MAE/RMSE/R² on 5-day forecasts that are uniformly better than 1-day forecasts, and every benchmark model exhibits the same anomaly. A second anomaly appears in the 60-day row of Table 11: MAPE collapses to 0.105% while MAE=28.73 and RMSE=36.02, which is mathematically inconsistent with the corn price range of 1789–3043 reported in Table 2. The author also states that long-term forecasts no longer use misaligned datasets because future information can be gathered, which strongly suggests information leakage (use of future values as inputs). Concerns are limited to the inferences drawn from tables and methodology text; raw data and code were not independently examined.

Verdict

🔴 Severe concerns — multiple indicators of data fabrication and/or information leakage. Confidence is moderate-to-high based on internal numerical inconsistencies and implausible reported results, but final determination requires raw data and code from the authors.

Key findings

Forecast horizon anomaly: the proposed CBA-X-PVW model and all five benchmark models show lower error metrics (MAE, RMSE) and higher R² on 5-day forecasts than on 1-day forecasts (Tables 7 and 11), contradicting basic time-series properties.
Implausible 10-day, 20-day, 40-day, and 60-day error trajectories, with non-monotonic behavior inconsistent with growing forecast uncertainty.
Mathematical inconsistency in Table 11, NO.7 row, 60-day forecast: MAPE = 0.105% is incompatible with MAE = 28.73 and the corn price range of 1789–3043 stated in Table 2 (implied true price ≈ 27,000).
Methodology statement in Section 5.4.3 explicitly abandons misaligned datasets for long-term forecasting because 'all current information can be gathered,' raising a strong suspicion of future-information leakage.

Evidence highlights

Table 7 (1-day, CBA-X-PVW): MAE = 10.82, RMSE = 12.96, R² = 0.9807.
Table 11 (5-day, CBA-X-PVW): MAE = 8.34, RMSE = 10.57, R² = 0.9871 — strictly better than 1-day for the same model.
Table 11, NO.7 MAPE trajectory: 0.461% (5-day) → 0.664% (10-day) → 0.749% (20-day) → 0.852% (40-day) → 0.105% (60-day), a discontinuous collapse.
Table 11, NO.7 60-day row: MAE = 28.73, RMSE = 36.02, MAPE = 0.105% — inconsistent with the 1789–3043 price band from Table 2.
DOI: 10.1007/s10479-025-06525-8.

Notes

All quantitative claims are reproduced exactly from the reported tables; no findings are invented.
The information-leakage interpretation depends on how the author operationalized 'current information' for long-horizon targets; the paper does not provide a rigorous definition.
Independent verification requires the authors' raw datasets, forecasting code, and the precise alignment/feature-construction procedure used for multi-step forecasts.
Marked uncertainty: without the underlying code, the anomalies are consistent with but not strictly proof of fabrication; alternative explanations (e.g., data leakage from rolling-window construction) must also be examined.
Recommended actions include requesting raw data and code, posting on PubPeer, notifying the journal editorial office (Annals of Operations Research), and reporting to the author's institution (Sun Yat-sen University Business School).