arxiv: 2605.02684 · v1 · submitted 2026-05-04 · 💻 cs.LG · physics.app-ph

Spectral Model eXplainer: a chemically-grounded explainability framework for spectral-based machine learning models

Jose Vinicius Ribeiro , Rafael Figueira Goncalves , Fabio Luiz Melquiades , Sylvio Barbon Junior This is my paper

Pith reviewed 2026-05-09 15:58 UTC · model grok-4.3

classification 💻 cs.LG physics.app-ph

keywords explainable AIspectral datachemometricsXAI frameworkmachine learning interpretabilityspectral zonespost-hoc explanationPCA

0 comments

The pith

The Spectral Model eXplainer attributes machine learning predictions on spectral data to expert-defined chemical zones rather than individual variables.

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

This paper introduces the Spectral Model eXplainer (SMX) to address the mismatch between standard XAI tools and the continuous, collinear nature of spectral data in chemometrics. Instead of scoring isolated spectral points, SMX operates on expert-specified zones by summarizing each with PCA, forming quantile-based logical predicates, scoring their relevance through targeted perturbations on stochastic subsamples, and aggregating the results into a directed weighted graph ranked by Local Reaching Centrality. A reconstruction step projects the predicate thresholds back onto the original spectrum so explanations can be overlaid directly on measured data. The approach matters for spectroscopy applications because domain experts can now inspect which chemically meaningful regions drive a classifier without first translating variable-level scores into zones. Evaluation on eight real XRF and gamma-ray datasets plus a synthetic benchmark shows the framework produces zone-level rankings that remain consistent across models.

Core claim

SMX is a post-hoc, global, model-agnostic framework that explains spectral classifiers by working directly with expert-informed spectral zones. Each zone is reduced to its principal components, logical predicates are created from quantile boundaries, predicate importance is estimated by measuring prediction change under stochastic perturbation, and the resulting rankings are aggregated in a directed graph whose nodes are zones and whose edges reflect co-occurrence strength; Local Reaching Centrality then yields a global zone ranking. Threshold spectrum reconstruction maps the active predicate boundaries back to the original wavelength or energy axis, producing synthetic spectra that can be相比

What carries the argument

Expert-informed spectral zones reduced by PCA to quantile predicates whose relevance is scored by perturbation and aggregated via directed-graph Local Reaching Centrality, with back-projection to threshold spectra.

If this is right

Spectral classifiers receive zone-level explanations that match the physical continuity and chemical interpretation used by spectroscopists.
The same expert zones can be reused across different models, enabling direct comparison of which chemical regions each model relies on.
Threshold spectra allow visual overlay of explanation boundaries on raw measurements in the original units.
Global rankings emerge from local predicate scores without requiring model-specific retraining.
The method applies uniformly to XRF and gamma-ray spectral datasets as shown in the eight real-world evaluations.

Where Pith is reading between the lines

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

Domain experts could iteratively refine zone definitions by inspecting whether SMX-highlighted zones align with known chemical signatures.
The zone-based approach might transfer to other continuous signals such as infrared or Raman spectra where physical regions carry semantic meaning.
If consistent zone importance patterns appear across related datasets, they could suggest candidate features for simpler, more interpretable models.
Integration with existing spectroscopy software would let analysts accept or reject model predictions based on chemical plausibility of the highlighted zones.

Load-bearing premise

Expert-defined spectral zones exist and are chemically meaningful enough that their PCA summaries and quantile predicates capture the model's actual decision boundaries without critical loss or bias from the subsampling step.

What would settle it

A spectral dataset in which the model's decisions depend on cross-zone interactions or narrow features outside the expert zones, causing SMX zone rankings to diverge systematically from SHAP or PFI importance when both are aggregated to the same zones.

Figures

Figures reproduced from arXiv: 2605.02684 by Fabio Luiz Melquiades, Jose Vinicius Ribeiro, Rafael Figueira Goncalves, Sylvio Barbon Junior.

**Figure 1.** Figure 1: Pipeline of the Spectral Model eXplainer (SMX). Given a preprocessed spectral matrix and a trained supervised model, SMX extracts global post-hoc explanations. quantifies how well the score captures the information content of the zone and is later used to weight edge importance in the graph. Although approximating to one component, PCA aggregation offers two key advantages over simpler aggregators (e.g.,… view at source ↗

**Figure 2.** Figure 2: Illustrative example of a threshold spectrum corresponding to a predicate defined on the Fe Kα zone from a XRF dataset covering two classes independent seeds R = {r1, . . . , rR}, and the final centrality score of the j-th predicate is taken as the j-th mean LRC(Pj ) = |R|−1 PR s=1 LRCrs (Pj ). This multi-seed averaging acts as a variance-reduction mechanism analogous to ensemble averaging in which indi… view at source ↗

**Figure 3.** Figure 3: Synthetic-generated spectra, colored by class. Vertical delimited areas indicate the defined spectral zones. a.u.: arbitrary units 5 Results and discussion 5.1 Spectral datasets and modeling results Figures 3 and 4 show the synthetic and a preprocessed XRF spectra (of the soil fertility dataset) to serve as a representative examples and to visually contextualize the defined spectral zones. The synthetic data ( view at source ↗

**Figure 4.** Figure 4: Poisson preprocessed spectra of the soil dataset, colored by class. Vertical delimited areas indicate the defined view at source ↗

**Figure 5.** Figure 5: Faithfulness analysis via progressive top-k masking. The plots show the per-k mean MAE (for PLS) and probability shift (for SVM and MLP) across the datasets (n=8). The legends indicate the mean area under the curve (AUC) encompassing the results for the first 9 most important zones for each method. The 9th zone was chosen as the maximum depth for visualization purposes, since it is the highest common dept… view at source ↗

**Figure 6.** Figure 6: Domain alignment analysis via agreement with expertdefined plausible zones. The plots show the per-zone mean cumulative agreement rate across the datasets (n=8) for each method. The legends indicate the mean area under the curve (AUC) encompassing the results for the first 9 most important zones for each method. The 9th zone was chosen as the maximum depth for visualization purposes, since it is the hig… view at source ↗

**Figure 7.** Figure 7: Stability analysis via pairwise RBO comparisons between the ranked predicate (SMX) and variable importance (PFI) lists obtained from 10 different random seeds for SMX and PFI, both with 4 internal repetitions. The histograms show the distribution of instability scores (1-RBO) across all seed pairs (n=45) averaged across the datasets (n=8) for each method. The mean ± standard deviation of the instability sc… view at source ↗

**Figure 9.** Figure 9: Violin plots showing the distribution of PC 1 scores for the spectral zones of the soil (XRF) dataset view at source ↗

**Figure 10.** Figure 10: Violin plots showing the distribution of PCA scores for the Ca kα zone. The horizontal dashed line indicates the threshold defined by the employed quantiles (q=[0.2, 0.4, 0.6, 0.8]) for drawing the predicates, computed on the complete training set each boundary. For well-separated classes such as eutrophic and dystrophic soils in the Ca kα zone, this visual comparison may suffice to support a preliminar… view at source ↗

**Figure 11.** Figure 11: Measured XRF spectra of eutrophic (Class A, red) and dystrophic (Class B, blue) soils overlaid with the threshold spectra τ spectrum (dotted curves, Eq. 11) of the three top-ranked SMX predicates for the PLS model. Each dotted curve represents the spectral profile of a sample lying exactly on the predicate boundary: spectra above it satisfy the > condition (predominantly eutrophic) and spectra below satis… view at source ↗

**Figure 12.** Figure 12: Ablation study on the number of bags hyperparameter employing the synthetic dataset. LRC: Local Reaching Centrality, Feat: Features, Back: Background 0.05 0.1 0.15 0.2 0.25 0.3 quantiles step size 2 4 6 Zone Rank MLP 0.05 0.1 0.15 0.2 0.25 0.3 quantiles step size 2 4 6 PLS 0.05 0.1 0.15 0.2 0.25 0.3 quantiles step size 2 4 6 SVM Zone Ranking Sensitivity: quantiles_step Feature 1 Feature 2 Feature 3 backgr… view at source ↗

**Figure 13.** Figure 13: Ablation study on the quantiles hyperparameter employing the synthetic dataset 19 view at source ↗

read the original abstract

Spectral-based machine learning models have been increasingly deployed in chemometrics and spectroscopy, where predictive accuracy is as important as explainability. Current employed eXplainable Artificial Intelligence (XAI) methods are largely adapted from tabular or generic multivariate domains, assigning relevance to isolated spectral variables rather than to the chemically meaningful spectral zones. Widely adopted tools such as SHapley Additive exPlanations (SHAP), Permutation Feature Importance (PFI), and Variable Importance in Projection scores (VIP) were not designed for the physical continuity and high collinearity of spectral data, and their variable-level outputs require post-hoc aggregation to recover zone-level information. This study introduces the Spectral Model eXplainer (SMX), a post-hoc, global, model-agnostic XAI framework that explains spectral classifiers through expert-informed spectral zones. SMX summarizes each zone via PCA, defines quantile-based logical predicates, estimates predicate relevance with perturbation in stochastic subsamples, and aggregates bag-wise rankings in a directed weighted graph summarized by Local Reaching Centrality. A key component is threshold spectrum reconstruction, which back-projects predicate boundaries to the original spectral domain in natural measurement units, enabling direct visual comparison with measured spectra. The method was evaluated on eight real spectral datasets (six based on X-ray Fluorescence--XRF and two based on Gamma-ray Spectrometry) and one synthetic benchmark with known gr

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.

Desk Editor's Note private letter to a colleague

SMX assembles a zone-based pipeline for spectral explanations but the abstract supplies no numbers or comparisons to show it works better than simpler aggregations.

read the letter

The new piece is the SMX pipeline: expert zones are summarized by PCA, turned into quantile predicates, scored by perturbation on stochastic subsamples, ranked through a directed graph with local reaching centrality, and back-projected as threshold spectra. That last step is practical because it returns explanations in the original wavelength units instead of abstract feature scores. The approach is aimed squarely at chemometrics users who already partition spectra into chemically meaningful regions and want the model output to respect those regions rather than treating every channel independently. The scope covers eight real XRF and gamma datasets plus a synthetic case, which is enough to demonstrate applicability across common spectral types. The description also avoids the usual post-hoc aggregation headaches that come with SHAP or VIP on collinear data. The central gap is the absence of any quantitative results, baseline comparisons, or stability checks in the text provided. Without those, it is impossible to tell whether the zone summaries and predicates actually recover the classifier's decision logic or whether the stochastic subsampling produces reproducible rankings. The assumption that expert zones are both available and sufficient is left untested; if the model relies on cross-zone correlations or fine structure inside a zone that PCA discards, the predicates will be incomplete. This work is for applied researchers in analytical chemistry who need zone-level explanations they can discuss with domain experts. A reader looking for a ready-to-use tool will find the steps clear enough to implement, but anyone wanting evidence of improved trust or accuracy will have to wait for the full experiments. I would send it to peer review because the framing addresses a recurring practical complaint in the field and the pipeline is reproducible from the description, even if the current write-up is thin on validation.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces the Spectral Model eXplainer (SMX), a post-hoc, global, model-agnostic XAI framework for spectral classifiers. SMX partitions spectra into expert-informed zones, summarizes each via PCA, defines quantile-based logical predicates, estimates predicate relevance through perturbation on stochastic subsamples, aggregates bag-wise rankings into a directed weighted graph summarized by Local Reaching Centrality, and reconstructs threshold spectra to project predicate boundaries back into the original spectral domain in physical units. The approach is evaluated on eight real datasets (six XRF, two gamma-ray spectrometry) plus one synthetic benchmark with known ground truth.

Significance. If the central claims hold, SMX would supply a chemically interpretable alternative to variable-level XAI tools (SHAP, PFI, VIP) by operating directly at the level of expert spectral zones and returning explanations in native measurement units. The threshold-spectrum reconstruction and graph-based aggregation via Local Reaching Centrality are distinctive technical contributions that could improve adoption in chemometrics and spectroscopy.

major comments (2)

[Method] Method description (around the SMX pipeline): the claim that zone-level PCA summaries plus quantile predicates faithfully encode the classifier's decision logic is load-bearing for the 'chemically-grounded' framing, yet the manuscript provides no ablation or counter-example testing cases where the model exploits cross-zone correlations or non-linear intra-zone structure discarded by linear PCA. Without such validation, the perturbation relevance scores and subsequent graph rankings may systematically misrepresent model behavior.
[Experiments] Evaluation section: the abstract and summary describe evaluation on eight real plus one synthetic dataset, but the provided text contains no quantitative metrics, baseline comparisons (e.g., zone-aggregated SHAP or VIP), variance estimates from the stochastic subsampling, or error analysis. This absence leaves the central claim of improved chemical grounding without verifiable numerical support.

minor comments (2)

[Abstract] The abstract is truncated mid-sentence ('known gr'); ensure the final version completes the description of the synthetic benchmark.
[Method] Notation for the directed graph and Local Reaching Centrality should be defined explicitly with an equation or pseudocode to avoid ambiguity when readers reconstruct the aggregation step.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below, indicating the revisions we will make.

read point-by-point responses

Referee: [Method] Method description (around the SMX pipeline): the claim that zone-level PCA summaries plus quantile predicates faithfully encode the classifier's decision logic is load-bearing for the 'chemically-grounded' framing, yet the manuscript provides no ablation or counter-example testing cases where the model exploits cross-zone correlations or non-linear intra-zone structure discarded by linear PCA. Without such validation, the perturbation relevance scores and subsequent graph rankings may systematically misrepresent model behavior.

Authors: We agree that validating the fidelity of zone-level PCA summaries and quantile predicates in capturing the model's full decision logic is essential to support the chemically-grounded framing. The current manuscript does not contain ablations or counter-examples that explicitly test scenarios involving cross-zone correlations or non-linear intra-zone structures discarded by linear PCA. To address this, we will add a dedicated subsection with synthetic benchmark experiments that introduce controlled cross-zone and non-linear effects, comparing SMX predicate relevance and graph rankings against the known ground-truth model behavior. This will clarify the framework's assumptions and limitations without altering the core method. revision: yes
Referee: [Experiments] Evaluation section: the abstract and summary describe evaluation on eight real plus one synthetic dataset, but the provided text contains no quantitative metrics, baseline comparisons (e.g., zone-aggregated SHAP or VIP), variance estimates from the stochastic subsampling, or error analysis. This absence leaves the central claim of improved chemical grounding without verifiable numerical support.

Authors: We acknowledge that the evaluation section as presented lacks the quantitative metrics, baseline comparisons, variance estimates, and error analysis needed to substantiate the claims. Although the manuscript outlines the datasets and qualitative aspects of the results, it does not report specific numerical values or statistical details. In the revision, we will expand the experiments section to include quantitative fidelity and stability metrics for SMX explanations, direct comparisons against zone-aggregated SHAP and VIP baselines, variance estimates derived from the stochastic subsampling procedure, and an accompanying error analysis. These additions will supply the verifiable numerical support referenced in the abstract. revision: yes

Circularity Check

0 steps flagged

SMX framework applies standard statistical primitives in a novel combination with no self-referential reductions.

full rationale

The paper introduces SMX as a post-hoc model-agnostic explainer that summarizes expert-defined spectral zones via PCA, constructs quantile predicates, estimates relevance via perturbation on stochastic subsamples, builds a directed graph of predicate rankings, and computes Local Reaching Centrality. These steps rely on well-known, externally defined operations (PCA, quantiles, perturbation importance, graph centrality) applied to the model's outputs rather than deriving new quantities that loop back to fitted parameters or self-citations. No equations reduce the final zone rankings or threshold spectra to inputs by construction, and the central claim of chemically-grounded explanations rests on the independent validity of expert zones and the faithfulness of the summarization, not on any internal tautology. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on domain assumptions about the utility of expert zones and the validity of the aggregation procedure rather than new physical postulates or fitted constants.

axioms (2)

domain assumption Expert-informed spectral zones capture chemically meaningful structure in the data.
Invoked when the method requires these zones as input and uses them to define predicates.
domain assumption Perturbation in stochastic subsamples yields stable relevance estimates for the predicates.
Underlies the estimation step that feeds the graph construction.

pith-pipeline@v0.9.0 · 5564 in / 1342 out tokens · 37177 ms · 2026-05-09T15:58:35.211165+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

43 extracted references · 9 canonical work pages

[1]

Explain- able artificial intelligence for spectroscopy data: a re- view.Pflügers Archiv - European Journal of Physiol- ogy, 477(4):603–615, 2025

Jhonatan Contreras and Thomas Bocklitz. Explain- able artificial intelligence for spectroscopy data: a re- view.Pflügers Archiv - European Journal of Physiol- ogy, 477(4):603–615, 2025. doi:10.1007/s00424-024- 02997-y. URL https://link.springer.com/ article/10.1007/s00424-024-02997-y

work page doi:10.1007/s00424-024- 2025
[2]

Peeking inside the black-box: A survey on explainable artifi- cial intelligence (xai).IEEE Access, 6:52138–52160,

Amina Adadi and Mohammed Berrada. Peeking inside the black-box: A survey on explainable artifi- cial intelligence (xai).IEEE Access, 6:52138–52160,
[3]

doi:10.1109/ACCESS.2018.2870052

work page doi:10.1109/access.2018.2870052 2018
[4]

Notions of explain- ability and evaluation approaches for explainable ar- tificial intelligence.Information Fusion, 76:89–106,

Giulia Vilone and Luca Longo. Notions of explain- ability and evaluation approaches for explainable ar- tificial intelligence.Information Fusion, 76:89–106,
[5]

doi:10.1016/j.inffus.2021.05.009

work page doi:10.1016/j.inffus.2021.05.009 2021
[6]

A unified approach to interpreting model predictions

Scott M Lundberg and Su-In Lee. A unified approach to interpreting model predictions. InAdvances in Neural Information Processing Systems, volume 30, 2017

2017
[7]

Why should I trust you?

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. “Why should I trust you?”: Explaining the predictions of any classifier. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1135–1144, 2016

2016
[8]

Grad-CAM: Visual explanations from deep networks via gradient-based localization

Ramprasaath R Selvaraju, Michael Cogswell, Ab- hishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual explanations from deep networks via gradient-based localization. InProceedings of the IEEE International Conference on Computer Vision, pages 618–626, 2017

2017
[9]

Learning deep features for discriminative localization

Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discriminative localization. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2921–2929, 2016

2016
[10]

Nir and mir spectroscopy for quick detection of the adulteration of cocoa content in chocolates.Food Chemistry, 349:129095, 2021

Ingrid Alves Santos, Daniele Gomes Conceição, Marília Borges Viana, Grazielly de Jesus Silva, Lean- dro Soares Santos, and Sibelli Passini Barbosa Ferrão. Nir and mir spectroscopy for quick detection of the adulteration of cocoa content in chocolates.Food Chemistry, 349:129095, 2021

2021
[11]

Support vector machines in tandem with infrared spectroscopy for geographical classification of green arabica coffee.LWT-Food Science and Tech- nology, 76:330–336, 2017

Evandro Bona, Izabele Marquetti, Jade Varaschim Link, Gustavo Yasuo Figueiredo Makimori, Vinícius da Costa Arca, André Luis Guimarães Lemes, Ju- liana Mendes Garcia Ferreira, Maria Brigida dos Santos Scholz, Patricia Valderrama, and Ronei Je- sus Poppi. Support vector machines in tandem with infrared spectroscopy for geographical classification of green a...

2017
[12]

Patricia Casarin, Franciele Leila Giopato Viell, Cintia Sorane Good Kitzberger, Luana Dalagrana Dos San- tos, Fábio Melquiades, and Evandro Bona. Determi- nation of the proximate composition and detection of adulterations in teff flours using near-infrared spec- troscopy.Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 334:125955, 2025

2025
[13]

Xrf and gamma-ray data fusion for pre- dicting key soil fertility attributes.Radiation Physics and Chemistry, 234:112750, 2025

Jose Vinicius Ribeiro, Joao Marcos Favaro Lopes, Avacir C Andrello, Jose Francirlei de Oliveira, Gra- ziela MC Barbosa, Rodrigo O Bastos, and Fabio Luiz Melquiades. Xrf and gamma-ray data fusion for pre- dicting key soil fertility attributes.Radiation Physics and Chemistry, 234:112750, 2025

2025
[14]

José Vinícius Ribeiro, Felipe Rodrigues dos Santos, José Francirlei de Oliveira, Graziela MC Barbosa, and Fábio Luiz Melquiades. Optimization of pxrf in- strumentation conditions and multivariate modeling in soil fertility attributes determination.Spectrochim- ica Acta Part B: Atomic Spectroscopy, 211:106835, 2024

2024
[15]

Soybean sorting based on protein content us- ing x-ray fluorescence spectrometry.Food Chemistry, 412:135548, 2023

Rachel Ferraz de Camargo, Tiago Rodrigues Tavares, Nicolas Gustavo da Cruz da Silva, Eduardo de Almeida, and Hudson Wallace Pereira de Car- valho. Soybean sorting based on protein content us- ing x-ray fluorescence spectrometry.Food Chemistry, 412:135548, 2023

2023
[16]

Low-cost spectroscopic devices with multivariate analysis applied to milk authenticity.Microchemical Journal, 181:107746, 2022

Diego Galvan, Carini Aparecida Lelis, Luciane Efft- ing, Fábio Luiz Melquiades, Evandro Bona, and Car- los Adam Conte-Junior. Low-cost spectroscopic devices with multivariate analysis applied to milk authenticity.Microchemical Journal, 181:107746, 2022

2022
[17]

Quantitative analysis of cadmium in rice roots based on libs and chemometrics methods.Environmental Sciences Eu- rope, 33(1):37, 2021

Wei Wang, Wenwen Kong, Tingting Shen, Zun Man, Wenjing Zhu, Yong He, and Fei Liu. Quantitative analysis of cadmium in rice roots based on libs and chemometrics methods.Environmental Sciences Eu- rope, 33(1):37, 2021

2021
[18]

Compositional analysis of copper and iron- based alloys using libs coupled with chemometric method.Analytical Sciences, 40(1):53–65, 2024

Vikas Gupta, Abhishekh Kumar Rai, Tejmani Ku- mar, Akash Tarai, G Manoj Kumar Gundawar, and AK Rai. Compositional analysis of copper and iron- based alloys using libs coupled with chemometric method.Analytical Sciences, 40(1):53–65, 2024

2024
[19]

Development of a rapid x-ray fluorescence method for protein determination in soybean grains.Food Chemistry, 473:143095, 2025

Rachel Ferraz de Camargo, Tiago Rodrigues Tavares, Felipe Rodrigues Dos Santos, and Hudson Wal- lace Pereira de Carvalho. Development of a rapid x-ray fluorescence method for protein determination in soybean grains.Food Chemistry, 473:143095, 2025

2025
[20]

Impact of calibration set size for predicting soil fertility attributes using local pxrf spectral libraries.Soil Advances, 3:100031, 2025

José Vinícius Ribeiro, Tiago Rodrigues Tavares, José Francirlei de Oliveira, Graziela MC Barbosa, and Fábio Luiz Melquiades. Impact of calibration set size for predicting soil fertility attributes using local pxrf spectral libraries.Soil Advances, 3:100031, 2025

2025
[21]

Variable selection by permutation applied in support vector regression models.Journal of Chemometrics, 36(10):e3444, 2022

Pedro HP da Cunha, Ellisson H de Paulo, Gabriely Silveira Folli, Márcia HC Nascimento, Mariana K Moro, and Paulo R Filgueiras. Variable selection by permutation applied in support vector regression models.Journal of Chemometrics, 36(10):e3444, 2022

2022
[22]

Interval partial least- squares regression (iPLS): A comparative chemo- metric study with an example from near-infrared spectroscopy.Applied Spectroscopy, 54(3):413–419,

Lars Nørgaard, Anders Saudland, Jacob Wag- ner, Jens Peter Nielsen, Lars Munck, and Søren Balling Engelsen. Interval partial least- squares regression (iPLS): A comparative chemo- metric study with an example from near-infrared spectroscopy.Applied Spectroscopy, 54(3):413–419,
[23]

doi:10.1366/0003702001949500

work page doi:10.1366/0003702001949500
[24]

Selec- tion of individual variables versus intervals of vari- ables in plsr.Journal of Chemometrics: A Journal of the Chemometrics Society, 24(2):45–56, 2010

Masoud Shariati-Rad and Masoumeh Hasani. Selec- tion of individual variables versus intervals of vari- ables in plsr.Journal of Chemometrics: A Journal of the Chemometrics Society, 24(2):45–56, 2010

2010
[25]

Decision predicate graphs: Enhancing interpretability in tree ensembles

Leonardo Arrighi, Luca Pennella, Gabriel Mar- ques Tavares, and Sylvio Barbon Junior. Decision predicate graphs: Enhancing interpretability in tree ensembles. InWorld Conference on Explainable Ar- tificial Intelligence, pages 311–332. Springer, 2024

2024
[26]

From anecdotal evidence to quantitative evaluation methods: A systematic review on evaluating explain- able ai.ACM Computing Surveys, 55(13s):1–42, 2023

Meike Nauta, Jan Trienes, Shreyasi Pathak, Elisa Nguyen, Michelle Peters, Yasmin Schmitt, Jörg Schlötterer, Maurice Van Keulen, and Christin Seifert. From anecdotal evidence to quantitative evaluation methods: A systematic review on evaluating explain- able ai.ACM Computing Surveys, 55(13s):1–42, 2023

2023
[27]

Explainable artificial in- telligence (xai) 2.0: A manifesto of open challenges and interdisciplinary research directions.Information Fusion, 106:102301, 2024

Luca Longo, Mario Brcic, Federico Cabitza, Jae- sik Choi, Roberto Confalonieri, Javier Del Ser, Ric- cardo Guidotti, Yoichi Hayashi, Francisco Herrera, Andreas Holzinger, et al. Explainable artificial in- telligence (xai) 2.0: A manifesto of open challenges and interdisciplinary research directions.Information Fusion, 106:102301, 2024

2024
[28]

A review of taxonomies of explainable artificial intelligence (xai) methods

Timo Speith. A review of taxonomies of explainable artificial intelligence (xai) methods. InProceedings of the 2022 ACM conference on fairness, account- ability, and transparency, pages 2239–2250, 2022

2022
[29]

6276–6282

Tim Miller. Explanation in artificial intelligence: In- sights from the social sciences.Artificial Intelligence, 267:1–38, 2019. doi:10.1016/j.artint.2018.07.007

work page doi:10.1016/j.artint.2018.07.007 2019
[30]

Hier- archy measure for complex networks.PloS one, 7(3): e33799, 2012

Enys Mones, Lilla Vicsek, and Tamás Vicsek. Hier- archy measure for complex networks.PloS one, 7(3): e33799, 2012

2012
[31]

Computer aided design of experiments.Technometrics, 11(1): 137–148, 1969

Ronald W Kennard and Larry A Stone. Computer aided design of experiments.Technometrics, 11(1): 137–148, 1969

1969
[32]

Timeshap: Explain- ing recurrent models through sequence perturbations

João Bento, Pedro Saleiro, André F Cruz, Mário AT Figueiredo, and Pedro Bizarro. Timeshap: Explain- ing recurrent models through sequence perturbations. InProceedings of the 27th ACM SIGKDD confer- ence on knowledge discovery & data mining, pages 2565–2573, 2021

2021
[33]

Groupsegment- shap: Shapley value explanations with group- 22 Spectral Model eXplainerA PREPRINT segment players for multivariate time series.arXiv preprint arXiv:2601.06114, 2026

Jinwoong Kim and Sangjin Park. Groupsegment- shap: Shapley value explanations with group- 22 Spectral Model eXplainerA PREPRINT segment players for multivariate time series.arXiv preprint arXiv:2601.06114, 2026

work page arXiv 2026
[34]

All models are wrong, but many are use- ful: Learning a variable’s importance by studying an entire class of prediction models simultaneously

Aaron Fisher, Cynthia Rudin, and Francesca Do- minici. All models are wrong, but many are use- ful: Learning a variable’s importance by studying an entire class of prediction models simultaneously. Journal of Machine Learning Research, 20(177):1– 81, 2019

2019
[35]

Pls: partial least squares projections to latent struc- tures

Svante Wold, Erik Johansson, Marina Cocchi, et al. Pls: partial least squares projections to latent struc- tures. In3D QSAR in Drug Design: Theory, Methods and Applications., pages 523–550. Kluwer ESCOM Science Publisher, 1993

1993
[36]

Kurz and J

Burkhard Beckhoff, Birgit Kanngiesser, Norbert Langhoff, Reiner Wedell, and Helmut Wolff.Hand- book of Practical X-Ray Fluorescence Analysis. Springer, Berlin, 2006. doi:10.1007/978-3-540- 36722-2

work page doi:10.1007/978-3-540- 2006
[37]

Marcel Dekker, New York, 2nd edition, 2001

René Van Grieken and Andrzej Markowicz.Hand- book of X-Ray Spectrometry. Marcel Dekker, New York, 2nd edition, 2001

2001
[38]

Wiley, Chichester, 2nd edition, 2008

Gordon Gilmore and John Hemingway.Practical Gamma-Ray Spectrometry. Wiley, Chichester, 2nd edition, 2008. doi:10.1002/9780470861981

work page doi:10.1002/9780470861981 2008
[39]

Knoll.Radiation Detection and Measure- ment

Glenn F. Knoll.Radiation Detection and Measure- ment. Wiley, New York, 4th edition, 2010

2010
[40]

Individual comparisons by ranking methods.Biometrics bulletin, 1(6):80–83, 1945

Frank Wilcoxon. Individual comparisons by ranking methods.Biometrics bulletin, 1(6):80–83, 1945

1945
[41]

ACM Trans

William Webber, Alistair Moffat, and Justin Zobel. A similarity measure for indefinite rankings.ACM Transactions on Information Systems (TOIS), 28(4): 1–38, 2010. doi:10.1145/1852102.1852106

work page doi:10.1145/1852102.1852106 2010
[42]

Sistema brasileiro de classificação de solos.Centro Nacional de Pesquisa de Solos: Rio de Janeiro, page 306, 2018

EMBRAPA. Sistema brasileiro de classificação de solos.Centro Nacional de Pesquisa de Solos: Rio de Janeiro, page 306, 2018

2018
[43]

Soil fertility basics.Soil science extension, North carolina state university, 22:46–59, 2010

Steven C Hodges. Soil fertility basics.Soil science extension, North carolina state university, 22:46–59, 2010. A Appendix A.1 Pseudocode and computational complexity This section provides the complete SMX implementation through the summarized Algorithms 1–4. The algorithm 1 encodes the zones decomposition and aggregation by PCA, the algorithm 2 encodes t...

2010