Integrity Review: "Improving the classification performance of microplastics by noise reduction and baseline correction of Raman spectra with a neural network-based algorithm"

Academic fraud report · Geng Detector

Summary

This report assigns a yellow (questionable) verdict to the paper published in Optics Express (DOI: 10.1364/OE.597337). Two issues were flagged. First, Section 3.4 states that ten independent training and evaluation runs were performed using diverse random seed initializations, yet Table 4 reports only single-point accuracy values (e.g., 96.90%, 55.70%) without any standard deviation or standard error. This omission prevents assessment of model stability and falls below the reporting standards expected of a top-tier journal. Second, the reference list contains multiple incomplete entries: Refs. [1] and [2] lack journal names (only volume/issue numbers such as 59(12) and 386(6720) are given), and Ref. [20] omits author names entirely. Additionally, several in-text figure references appear broken (e.g., "(Fig.(a))", "(Fig.)"), which may reflect PDF extraction artifacts but could also indicate editorial sloppiness. Combined with the unusually short publication window (submission March 17, publication April 30), these findings raise legitimate but non-conclusive concerns about reporting rigor.

Verdict

🟡 Questionable. The paper exhibits reporting deficiencies that fall below expected standards for a journal of this caliber, though no definitive evidence of misconduct has been identified. The two flagged issues are reproducible from the published text.

Key findings

Missing variability statistics for repeated experiments: The authors claim ten independent runs with varied random seeds but report only single mean values without standard deviations or confidence intervals.
Incomplete reference list: Multiple references lack journal names, and at least one reference is missing author information entirely.
Broken in-text figure references: Several figure citations are malformed (e.g., "(Fig.(a))", "(Fig.)"), which may originate from PDF extraction but could also reflect formatting negligence.
Atypical publication timeline: The interval between submission (March 17) and publication (April 30) is unusually short, approximately six weeks.

Evidence highlights

Table 4 / Section 3.4 (Statistical reporting): Exact text — "independent training and evaluation runs were performed ten times using diverse global random seed initializations." Reported values include 96.90% and 55.70% as single numbers with no associated dispersion measure. Deep learning models are known to be seed-sensitive, making variance reporting essential.
Reference list (Citation completeness):
Ref [1]: missing journal name; only "59(12)" provided.
Ref [2]: missing journal name; only "386(6720)" provided.
Ref [20]: missing authors; entry begins directly with "dual-tree complex wavelet transform,".
In-text citations (Formatting): Section 3.1 contains "(Fig.(a))" and "(Fig.)"; Section 3.2 contains "(Fig.(a))".
Timeline: Submission 17 March 2026; publication 30 April 2026 (DOI: 10.1364/OE.597337).

Notes

The missing standard deviations are the most substantive concern and are easily verifiable by requesting raw run-by-run data from the corresponding author.
The malformed figure references may be artifacts of automated PDF text extraction rather than actual typographical errors in the typeset article; this should be checked against the publisher's HTML or official PDF.
No image manipulation, data fabrication, or plagiarism indicators were detected in this review.
The unusually fast turnaround may reflect efficient handling but warrants scrutiny of editorial process.
All findings are based solely on the published article; conclusions are limited to reporting quality rather than research integrity per se.
Recommended action: contact authors to request the per-run accuracy values for the ten random seed experiments and the full formatted reference list.