pith. sign in

arxiv: 1906.10242 · v1 · pith:SF72BBB5new · submitted 2019-06-24 · 💻 cs.LG · eess.SP· stat.ML

Multi-label Classification with Optimal Thresholding for Multi-composition Spectroscopic Analysis

Luyun Gan , Brosnan Yuen , Tao Lu This is my paper

Pith reviewed 2026-05-25 17:14 UTC · model grok-4.3

classification 💻 cs.LG eess.SPstat.ML
keywords multi-label classificationoptimal thresholdinginfrared spectroscopygas identificationneural networksspectroscopic analysismulti-composition detectionbinary relevance
0
0 comments X

The pith

Multi-label neural networks with optimal thresholding outperform conventional binary relevance methods for identifying multiple gases in infrared spectra when signal quality and training data are sufficient.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a multi-label neural network approach paired with optimal thresholding to detect multiple gas species from their combined infrared absorption spectra in mixed environments. It tests this on synthesized spectral data and reports better results than the standard binary relevance plus partial least squares discriminant analysis approach, but only when signal-to-noise ratio is high enough and enough training examples are present. A sympathetic reader would care because the method could allow direct analysis of overlapping signals without first isolating individual components, which matters for monitoring gas mixtures in settings like environmental sensing or industrial safety.

Core claim

The authors establish that multi-label classification with optimal thresholding applied to neural networks identifies gas species among a multi-gas mixture in a cluttered environment using infrared absorption spectroscopy, and that this outperforms conventional binary relevance partial least squares discriminant analysis when signal-to-noise ratio and training sample size are sufficient, as shown on synthesized spectral datasets.

What carries the argument

Multi-label neural networks with optimal thresholding, which assign multiple class labels at once and tune decision thresholds to handle simultaneous gas detections in one spectrum.

If this is right

  • Enables direct multi-gas identification from a single combined spectrum without physical separation.
  • Delivers higher accuracy than binary relevance partial least squares discriminant analysis under adequate signal-to-noise ratio and sample size.
  • Supports spectroscopic analysis tasks in cluttered or mixed environments.
  • Depends on the availability of sufficient training data and clean signals for its performance gain.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could be applied to other spectroscopic modalities or sensor types beyond infrared absorption.
  • Validation against real experimental mixtures rather than only synthesized data would test whether the outperformance transfers outside the training conditions.
  • Pairing the approach with noise-robust preprocessing might extend its usefulness to lower signal-to-noise ratio regimes.

Load-bearing premise

The synthesized spectral datasets accurately represent real-world multi-gas mixtures in cluttered environments.

What would settle it

A side-by-side test on measured experimental spectra from actual multi-gas mixtures that shows the neural network method loses its reported advantage over binary relevance partial least squares discriminant analysis.

Figures

Figures reproduced from arXiv: 1906.10242 by Brosnan Yuen, Luyun Gan, Tao Lu.

Figure 2
Figure 2. Figure 2: (a) Percentage of cumulative explained variance vs. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: (a) FNN-OT training and testing procedure. (b) A [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Learning curves of FNN-OT without dropout (blue), [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Micro-averaged (a) precision, (b) recall and (c) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Minimum detectable concentration (Cmin) of nine gases. IV. RESULTS AND DISCUSSIONS A. Hyper parameter tuning In our research, we use TensorFlow to implement our FNN￾OT and Adam as our optimizer. In first step we tune hyper￾parameters such as dropout rate and training sample size of the FNN-OT model with the SNR=30 dB data set. 1) Dropout: In order to tune the hyper-parameters for dropout, a grid search has… view at source ↗
Figure 5
Figure 5. Figure 5: As shown, both FNN-OT and FNN consistently show [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Micro-averaged (a) precision, (b) recall and (c) [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Joint distribution of two normal distributed random variables [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
read the original abstract

In this paper, we implement multi-label neural networks with optimal thresholding to identify gas species among a multi gas mixture in a cluttered environment. Using infrared absorption spectroscopy and tested on synthesized spectral datasets, our approach outperforms conventional binary relevance - partial least squares discriminant analysis when signal-to-noise ratio and training sample size are sufficient.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper proposes multi-label neural networks equipped with optimal thresholding for identifying multiple gas species from infrared absorption spectra in mixtures. It reports that this approach outperforms binary relevance partial least squares discriminant analysis on synthesized spectral datasets when signal-to-noise ratio and training sample size are sufficient.

Significance. If the performance gains are robust, the method could offer a practical improvement for multi-composition spectroscopic identification tasks. The work is motivated by real-world cluttered environments, but its current evaluation is confined to synthetic linear mixtures, which limits the strength of the applicability claim.

major comments (1)
  1. [Abstract and Results] The central performance claim (outperformance over binary relevance PLS-DA) rests exclusively on synthesized spectral datasets formed as linear combinations of reference spectra plus additive noise. No results on measured FTIR spectra from actual multi-gas mixtures are presented, which directly undermines the claim of utility in cluttered real-world environments. This is load-bearing for the paper's motivation and conclusions.
minor comments (1)
  1. [Abstract] The abstract states the method is 'tested on synthesized spectral datasets' but provides no quantitative details on dataset size, SNR levels, number of gas species, or error bars; these should be summarized in the abstract or a dedicated table.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their constructive comments. We address the major comment below.

read point-by-point responses

  1. Referee: [Abstract and Results] The central performance claim (outperformance over binary relevance PLS-DA) rests exclusively on synthesized spectral datasets formed as linear combinations of reference spectra plus additive noise. No results on measured FTIR spectra from actual multi-gas mixtures are presented, which directly undermines the claim of utility in cluttered real-world environments. This is load-bearing for the paper's motivation and conclusions.

    Authors: We agree that all reported results use synthetic datasets formed as linear combinations plus noise, as explicitly stated in the abstract and throughout the manuscript. This controlled generation permits precise variation of the number of species, concentrations, and SNR to enable rigorous method comparison where ground truth is known. We acknowledge that experimental validation on measured FTIR spectra from real multi-gas mixtures would provide stronger support for applicability in cluttered environments. Because such data collection lies outside the present study, we will revise the abstract, introduction, and conclusions to clarify the synthetic scope of the claims and to position real-world validation as future work. revision: partial

standing simulated objections not resolved
  • Results on measured FTIR spectra from actual multi-gas mixtures

Circularity Check

0 steps flagged

Empirical ML application on synthetic spectra exhibits no circular derivation

full rationale

The manuscript presents an applied machine-learning method (multi-label NN with optimal thresholding) evaluated via direct empirical comparison against BR-PLS-DA on synthesized linear-mixture spectra. No first-principles derivation, uniqueness theorem, or predictive equation is claimed; performance metrics are obtained by training and testing on the same class of synthetic data without any step that reduces a reported result to a fitted parameter by construction or to a self-citation chain. The work is therefore self-contained as an engineering demonstration.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities; ledger is empty by necessity.

pith-pipeline@v0.9.0 · 5573 in / 888 out tokens · 27096 ms · 2026-05-25T17:14:17.782650+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages · 2 internal anchors

  1. [1]

    Neural networks and the classification of mineralogical samples using X-ray spectra,

    M. Gallagher and P. Deacon, “Neural networks and the classification of mineralogical samples using X-ray spectra,” inProceedings of the 9th In- ternational Conference on Neural Information Processing. ICONIP’02. , vol. 5. pp. 2683–2687. IEEE, 2002,

    work page 2002

  2. [2]

    TDLAS-based detection of dissolved methane in power transformer oil and field application,

    J. Jiang, M. Zhao, G.-M. Ma, H.-T. Song, C.-R. Li, X. Han, and C. Zhang, “TDLAS-based detection of dissolved methane in power transformer oil and field application,” IEEE Sensors Journal , vol. 18, no. 6, pp. 2318–2325, 2018. 8

    work page 2018

  3. [3]

    Rapid and real-time analysis of volatile compounds released from food using infrared and laser spectroscopy,

    D. Dong, L. Jiao, C. Li, and C. Zhao, “Rapid and real-time analysis of volatile compounds released from food using infrared and laser spectroscopy,” TrAC Trends in Analytical Chemistry , 2018

    work page 2018

  4. [4]

    Real-time measurement of soil attributes using on-the- go near infrared reflectance spectroscopy,

    C. D. Christy, “Real-time measurement of soil attributes using on-the- go near infrared reflectance spectroscopy,” Computers and Electronics in Agriculture, vol. 61, no. 1, pp. 10–19, 2008

    work page 2008

  5. [5]

    TDLAS detection of propane/butane gas mixture by using reference gas absorption cells and partial least square approach,

    Y . Wang, Y . Wei, T. Liu, T. Sun, and K. T. Grattan, “TDLAS detection of propane/butane gas mixture by using reference gas absorption cells and partial least square approach,” IEEE Sensors Journal , vol. 18, no. 20, pp. 8587–8596

    work page

  6. [6]

    Identification and classification of organic and inorganic components of particulate matter via raman spectroscopy and chemometric approaches,

    W. Schumacher, M. K ¨uhnert, P. R ¨osch, and J. Popp, “Identification and classification of organic and inorganic components of particulate matter via raman spectroscopy and chemometric approaches,” Journal of Raman Spectroscopy , vol. 42, no. 3, pp. 383–392, 2011

    work page 2011

  7. [7]

    Explanatory analysis of spectroscopic data using ma- chine learning of simple, interpretable rules,

    R. Goodacre, “Explanatory analysis of spectroscopic data using ma- chine learning of simple, interpretable rules,” Vibrational Spectroscopy, vol. 32, no. 1, pp. 33–45, 2003

    work page 2003

  8. [8]

    An evaluation of statistical approaches to text categorization,

    Y . Yang, “An evaluation of statistical approaches to text categorization,” Information retrieval, vol. 1, no. 1-2, pp. 69–90, 1999

    work page 1999

  9. [9]

    Boostexter: A boosting-based system for text categorization,

    R. E. Schapire and Y . Singer, “Boostexter: A boosting-based system for text categorization,” Machine Learning, vol. 39, no. 2-3, pp. 135–168, 2000

    work page 2000

  10. [10]

    Multi-label classification: An overview,

    G. Tsoumakas and I. Katakis, “Multi-label classification: An overview,” International Journal of Data Warehousing and Mining , vol. 3, no. 3, 2006

    work page 2006

  11. [11]

    Multi-label learning: A review of the state of the art and ongoing research,

    E. Gibaja and S. Ventura, “Multi-label learning: A review of the state of the art and ongoing research,” Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery , vol. 4, no. 6, pp. 411–444, 2014

    work page 2014

  12. [12]

    Discriminative methods for multi-labeled classification,

    S. Godbole and S. Sarawagi, “Discriminative methods for multi-labeled classification,”Advances in Knowledge Discovery and Data Mining , pp. 22–30, 2004

    work page 2004

  13. [13]

    Multilabel text classification for automated tag suggestion,

    I. Katakis, G. Tsoumakas, and I. Vlahavas, “Multilabel text classification for automated tag suggestion,” ECML PKDD Discovery Challenge , vol. 75, 2008

    work page 2008

  14. [14]

    Random k-labelsets: An ensemble method for multilabel classification,

    G. Tsoumakas and I. Vlahavas, “Random k-labelsets: An ensemble method for multilabel classification,” in European Conference on Ma- chine Learning. Springer, pp. 406–417, 2007

    work page 2007

  15. [15]

    Classifier chains for multi-label classification,

    J. Read, B. Pfahringer, G. Holmes, and E. Frank, “Classifier chains for multi-label classification,”Machine Learning and Knowledge Discovery in Databases, pp. 254–269, 2009

    work page 2009

  16. [16]

    Knowledge discovery in multi-label pheno- type data,

    A. Clare and R. D. King, “Knowledge discovery in multi-label pheno- type data,” in European Conference on Principles of Data Mining and Knowledge Discovery. Springer, pp. 42–53, 2001

    work page 2001

  17. [17]

    A k-nearest neighbor based algorithm for multi-label classification,

    M.-L. Zhang and Z.-H. Zhou, “A k-nearest neighbor based algorithm for multi-label classification,” in 2005 IEEE International Conference on Granular Computing , vol. 2. pp. 718–721, 2005

    work page 2005

  18. [18]

    Multi-label Classification using Labels as Hidden Nodes

    J. Read and J. Hollm ´en, “Multi-label classification using labels as hidden nodes,” arXiv preprint arXiv:1503.09022 , 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  19. [19]

    Multilabel neural networks with applica- tions to functional genomics and text categorization,

    M.-L. Zhang and Z.-H. Zhou, “Multilabel neural networks with applica- tions to functional genomics and text categorization,” IEEE Transactions on Knowledge and Data Engineering , vol. 18, no. 10, pp. 1338–1351, 2006

    work page 2006

  20. [20]

    Large- scale multi-label text classification-revisiting neural networks,

    J. Nam, J. Kim, E. L. Menc ´ıa, I. Gurevych, and J. F ¨urnkranz, “Large- scale multi-label text classification-revisiting neural networks,” in Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, pp. 437–452, 2014

    work page 2014

  21. [21]

    A unified architecture for natural language processing: Deep neural networks with multitask learning,

    R. Collobert and J. Weston, “A unified architecture for natural language processing: Deep neural networks with multitask learning,” in Proceed- ings of the 25th International Conference on Machine Learning . ACM, pp. 160–167, 2008

    work page 2008

  22. [22]

    Deep Convolutional Ranking for Multilabel Image Annotation

    Y . Gong, Y . Jia, T. Leung, A. Toshev, and S. Ioffe, “Deep con- volutional ranking for multilabel image annotation,” arXiv preprint arXiv:1312.4894, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  23. [23]

    CNN- RNN: A unified framework for multi-label image classification,

    J. Wang, Y . Yang, J. Mao, Z. Huang, C. Huang, and W. Xu, “CNN- RNN: A unified framework for multi-label image classification,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2285–2294, 2016

    work page 2016

  24. [24]

    The HITRAN 2012 Molecular Spectroscopic Database,

    L. S. Rothman, I. E. Gordon, Y . Babikov, A. Barbe, D. C. Benner, P. F. Bernath, M. Birk, L. Bizzocchi, V . Boudon, L. R. Brown et al. , “The HITRAN 2012 Molecular Spectroscopic Database,” Journal of Quantitative Spectroscopy and Radiative Transfer , vol. 130, pp. 4–50, 2013

    work page 2012

  25. [25]

    Dropout: a simple way to prevent neural networks from over- fitting,

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhut- dinov, “Dropout: a simple way to prevent neural networks from over- fitting,”The Journal of Machine Learning Research , vol. 15, no. 1, pp. 1929–1958, 2014

    work page 1929

  26. [26]

    Principal components analysis (PCA),

    S. M. Holland, “Principal components analysis (PCA),” Department of Geology, University of Georgia, Athens, GA , pp. 30 602–2501, 2008

    work page 2008

  27. [27]

    C. S. Allred. Partially correlated uniformly distributed random numbers. [Online]. Available: https://medium.com/capital-one-tech/ partially-correlated-uniformly-distributed-random-numbers-5ce82486b68a

    work page

  28. [28]

    Binary relevance efficacy for multilabel classification,

    O. Luaces, J. D ´ıez, J. Barranquero, J. J. del Coz, and A. Bahamonde, “Binary relevance efficacy for multilabel classification,” Progress in Artificial Intelligence, vol. 1, no. 4, pp. 303–313, Dec 2012

    work page 2012

  29. [29]

    A machine learning application for classification of chemical spectra,

    M. G. Madden and T. Howley, “A machine learning application for classification of chemical spectra,” in Applications and Innovations in Intelligent Systems XVI . Springer, pp. 77–90, 2009

    work page 2009

  30. [30]

    Partial least-squares regression: A tutorial,

    P. Geladi and B. R. Kowalski, “Partial least-squares regression: A tutorial,” Analytica Chimica Acta , vol. 185, pp. 1–17, 1986

    work page 1986

  31. [31]

    Classification of raw milk by infrared spectroscopy (FTIR) and chemometric,

    M. Elbassbasi, F. Kzaiber, G. Ragno, and A. Oussama, “Classification of raw milk by infrared spectroscopy (FTIR) and chemometric,” Journal of Scientific Speculations and Research , vol. 1, no. 2, pp. 28–33, 2010

    work page 2010

  32. [32]

    Classification and quality control of lubricating oils by infrared spectroscopy and chemometric,

    A. Hirri, M. Bassbasi, and A. Oussama, “Classification and quality control of lubricating oils by infrared spectroscopy and chemometric,” Int. J. Adv. Technol. Eng. Res. , vol. 3, pp. 59–62, 2013

    work page 2013

  33. [33]

    FTIR spectroscopy and PLS-DA classification and prediction of four commercial grade virgin olive oils from Morocco,

    A. Hirri, M. Bassbasi, S. Platikanov, R. Tauler, and A. Oussama, “FTIR spectroscopy and PLS-DA classification and prediction of four commercial grade virgin olive oils from Morocco,” Food Analytical Methods, vol. 9, no. 4, pp. 974–981, 2016

    work page 2016