Data Integrity Concerns in 'SD-IoTR: an SDN-based Internet of Things reprogramming framework' (IET Networks, 2020)

Academic fraud report · Geng Detector

Summary

This report flags the paper 'SD-IoTR: an SDN-based Internet of Things reprogramming framework' by Dang Huynh-Van and Quan Le-Trung (IET Networks, 2020, DOI: 10.1049/iet-net.2019.0223) as highly suspicious based on internal numerical inconsistencies rather than image-based evidence. The central issue lies in Table 4 (Page 312): reported average bytes-per-second values cannot be reproduced from the listed total bytes and elapsed seconds. For Version 1, 270,768 Bytes / 101 s equals roughly 2680.8 Bytes/s, but the table reports 2663.6; for Version 3, 246,326 Bytes / 89 s equals about 2767.7 Bytes/s, but the table reports 2788.037. A secondary concern is rhetorical inconsistency in Section 4.1.2: prose claims CPU usage is 'just 26%', while a quoted confidence interval of 25.8–27.3% implies a mean near 26.55%, contradicting the narrative. Confidence is moderate; the arithmetic discrepancies are reproducible from the published numbers, but the underlying cause (typo, unit mismatch, or fabrication) cannot be definitively determined from the text alone. Figures 4–8 could not be visually inspected.

Verdict

🟠 Highly suspicious. Internal arithmetic and textual inconsistencies in reported experimental results warrant formal inquiry, though the evidence does not yet establish systematic fraud.

Key findings

Table 4 (Page 312) reports average bytes/s values that do not match the listed total bytes divided by elapsed seconds.
Version 1: 270,768 Bytes over 101 s ≈ 2680.8 Bytes/s, but the table reports 2663.6 Bytes/s.
Version 3: 246,326 Bytes over 89 s ≈ 2767.7 Bytes/s, but the table reports 2788.037 Bytes/s.
Version 3's average bytes/s is reported to three decimal places (2788.037), which is unusually precise for TinyOS-style network throughput measurements and inconsistent with the integer-granularity packet sizes elsewhere in the same table (e.g., 112, 113 bytes).
Section 4.1.2 (Page 312) prose states CPU utilisation is 'just 26%', while a confidence interval of 25.8–27.3% is then quoted; the midpoint (~26.55%) does not align with the narrative anchor of 26%.

Evidence highlights

Location: Table 4, Page 312. Numerical inconsistency between reported throughput and derivable throughput from total bytes and duration.
Location: Section 4.1.2, Page 312. Quantitative (95% CI) and qualitative ('just 26%') descriptions of CPU usage are not internally consistent.
Figures 4–8 (system architecture, CPU/memory traces) were not available for visual inspection; image-based checks were therefore not performed. Given the numerical issues, the accompanying performance curves should be examined in any follow-up.

Notes

All quantitative claims in this report are derived directly from the values printed in Table 4 and Section 4.1.2 of the published paper; no external data has been introduced.
The arithmetic discrepancies are reproducible and indicate either (a) transcription/unit errors in table preparation, (b) post-hoc smoothing or back-calculation of figures, or (c) fabricated metrics. The text alone cannot distinguish between these possibilities.
Recommended next steps: request the raw dstat/system logs and the script used to generate Table 4 from the authors; raise the specific Version 1 and Version 3 calculations on PubPeer; defer a journal-level complaint until the authors' response is evaluated.
DOI: 10.1049/iet-net.2019.0223