Pith. sign in

REVIEW 3 major objections 4 minor 23 references

Extending Decision Predicate Graphs for Comprehensive Explanation of Isolation Forest

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Turning an Isolation Forest into a graph of direction-annotated predicates, scored by the Inlier-Outlier Propagation Score, yields a global explanation of which features and in which direction drive outlier classification.

desk verdict A clearly specified, code-public XAI extension for Isolation Forest that loses the threshold information its 'boundary' claims depend on; worth a revise-and-resubmit, not a desk reject. read the letter →

arxiv 2505.04019 v1 pith:R7JYOYF5 submitted 2025-05-06 cs.AI cs.LG

Matteo Ceschin , Leonardo Arrighi , Luca Longo , Sylvio Barbon Junior This is my paper · ORCID
classification cs.AIcs.LG
keywords ExplainableAIIsolationForestDecisionPredicateGraphglobalexplanationoutlierdetectionIOP-Scorefeatureimportancemodelinterpretability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that Isolation Forest, an ensemble outlier detector, can be given a global explanation by converting its trees into a Decision Predicate Graph. Each graph node is a predicate pairing a feature with a direction ($\le$ or $>$), with split values removed so predicates can be compared across trees. The paper adds the Inlier-Outlier Propagation Score, which measures each predicate's tendency to push samples toward the outlier or inlier class. If the method works, a user can see which features separate anomalies, in which direction the separation goes, and whether a feature acts directly or only through longer decision paths. This matters because existing explanations of Isolation Forest mostly reduce to feature-importance vectors, which say which features matter but not how the model uses them.

What carries the argument

The central object is the Decision Predicate Graph (DPG) specialized to Isolation Forest. Each internal split rule of an isolation tree is first written as a triple (feature, sign, value), then the value is deleted so nodes become pairs such as $F_0 >$ or $TSH \le$, making predicates comparable across trees that randomly chose different thresholds. Nodes are connected by directed edges weighted by how often training samples satisfy the two predicates consecutively, with outlier and inlier frequencies reweighted to offset class imbalance. The Inlier-Outlier Propagation Score is defined per node as $(f_i(v)-f_o(v))/f_{in}(v)$, where $f_i$ and $f_o$ are the weighted frequencies of transitions from $v$ toward inlier and outlier classes and $f_{in}$ is the total weighted frequency entering $v$. This score is what turns the graph from a visualization into a quantitative statement about which predicates are discriminative and in which direction.

What would settle it

Train two Isolation Forests on the same data with different random threshold ranges so the same (feature, direction) predicates occur with matching frequencies while thresholds sit in very different places; if the DPGs and IOP-Scores are nearly identical but the models disagree on which points are outliers, the deleted threshold values carry information the explanation omits.

Watch

Extended reading notes

Core claim

The central claim is that representing a trained Isolation Forest as a weighted directed graph of direction-annotated feature predicates yields a global explanation of the model's outlier logic. Predicates of the form (feature, sign) become nodes; an edge from one predicate to the next records how often training samples satisfy them consecutively while traversing the trees, with frequencies reweighted so the rare outlier class is not drowned out by inliers. The IOP-Score for a node is the normalized difference between the frequency of transitions heading to the inlier class and the frequency heading to the outlier class; a score near $-1$ marks a predicate that strongly characterizes outliers, a score near $+1$ marks an inlier-defining predicate, and zero marks a neutral one. The paper argues that the graph plus these scores shows the features that isolate anomalies, the direction of the constraints (outliers above or below a split), and whether a predicate separates outliers directly or only when combined with further splits.

Load-bearing premise

The claim depends on the idea that dropping the actual threshold values from each split leaves a graph that still faithfully represents how the model isolates outliers.

Editorial extensions

If this is right

  • Users of an Isolation Forest can inspect one graph to see which features the model relies on to mark points as outliers, rather than only a ranked list of importances.
  • The sign attached to each predicate reveals the direction of the boundary, for instance that high TSH values, not low ones, drive the outlier class in the Annthyroid example.
  • Because edges carry reweighted frequencies, the graph also shows whether a predicate separates outliers immediately or only as part of a longer chain of conditions.
  • The IOP-Score gives a bounded, comparable value for every predicate, letting practitioners tell which features are discriminative for anomalies and which are neutral.
  • As a global explanation of the whole ensemble, the method supports auditing and fine-tuning of Isolation Forest rather than explaining only individual predictions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • My inference: the same predicate-pair abstraction could be applied to other tree ensembles whose splits use random thresholds, wherever a global explanation of boundary direction is wanted.
  • My inference: because split values are discarded, the same predicate pair can merge splits with very different thresholds; whether this loses fidelity is testable by comparing graphs from models trained with different threshold ranges.
  • My inference: the class-reweighting scheme assumes the model's own outlier and inlier labels are reliable, so noisy labels would propagate into the edge weights and the IOP-Score.
  • My inference: the IOP-Score could also work as an unsupervised feature-ranking signal, with predicates near $-1$ flagging candidate anomaly-relevant variables.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript proposes a post-hoc, global explanation method for Isolation Forest (iForest) based on Decision Predicate Graphs (DPG). The authors extract predicate triples (feature, operator, split value) from the internal nodes of iTrees, then delete the split values, keeping only (feature, direction) pairs. They construct a weighted directed graph over these predicates, introduce a class-based frequency weighting scheme, and define the Inlier-Outlier Propagation Score (IOP-Score) to quantify each predicate's tendency to lead toward the inlier or outlier class. The approach is demonstrated on two synthetic datasets and the Annthyroid benchmark, with qualitative visual and score-based interpretations. The central claim is that the method provides a 'comprehensive global explanation' of iForest, including 'feature boundaries' and the direction in which features drive outlier identification.

Significance. The paper addresses a real gap: most iForest explanation methods produce feature-importance vectors rather than structural descriptions of the model's logic. If the proposed graph representation were faithful, it would offer a new, visually interpretable modality for understanding iForest. The method is clearly specified, the equations for the weighting and IOP-Score are explicit, and the authors state that the full implementation is publicly available, which supports reproducibility. However, the central promise of explaining 'feature boundaries' is not actually delivered by the representation, because split values are deliberately removed. The experimental section is entirely qualitative and does not compare against existing methods such as SHAP, DIFFI, or ExIFFI, despite the introduction framing the work as overcoming their limitations. The strength of the graph-based idea is real, but the evidence provided is insufficient to support the breadth of the claimed contribution.

major comments (3)
  1. [Section 3.1, abstract, contribution list] The removal of split values from each predicate triple, reducing (f,σ,v) to (f,σ), is load-bearing for the paper's central claim of providing 'insights into decision boundaries' and 'details on feature boundaries for both inliers and outliers.' Without the threshold value, the representation cannot distinguish between, for example, 'X <= 10' and 'X <= 100', which are completely different decision regions in the original iForest. A concrete failure case: if inliers are clustered near 0 and near 1 and an outlier lies near 0.5, different random splits will sometimes produce 'X <= v' and sometimes 'X > v' as the predicate that isolates the outlier; the graph would contain both directions with comparable frequencies, IOP-Scores near zero, and no way to tell the user that the outlier is in the middle gap. Thus the proposed representation, as defined, cannot substantiate the advertised 'feature boundary' explanations.
  2. [Section 3.1, Algorithm 1, line 10] The step 'Remove paths of predicates exceeding treed max for outliers' rests on the statement that 'iForest classifies observations that reach the maximum tree depth as inliers.' This is not correct for the standard Isolation Forest. In the scoring formula given in Section 2.1, the anomaly score is s(x,n)=2^{−E(h(x))/c(n)}, where E(h) is averaged over all trees; an instance whose path length equals the maximum depth can still have s(x,n)>0.5, so it would be classified as an outlier. For example, with n=200 and default max depth 8, c(200)≈9.75 and s=2^{-8/9.75}≈0.566>0.5. Discarding all outlier traversals that reach the maximum depth therefore drops real outlier paths and biases the graph toward shallow, easy isolations, misrepresenting the model's actual decision process.
  3. [Section 4, Section 5] The experimental evaluation is qualitative only: the authors inspect node colors, edge thicknesses, and IOP-Score tables on three datasets, and no quantitative comparison is made with SHAP, DIFFI, ExIFFI, or any other baseline, nor are fidelity or faithfulness metrics reported. The manuscript itself acknowledges in Section 5 that 'a more in-depth comparison with these techniques is necessary to establish the specific advantages and trade-offs of DPG.' Since the framing of the paper is that DPG overcomes the limitations of existing feature-importance methods, the absence of any comparative or quantitative validation leaves the central explanatory claim unsubstantiated.
minor comments (4)
  1. [Front matter] The keywords list contains 'Food quality · Food engineering', which is unrelated to the paper's content and appears to be a copy-paste error from another manuscript.
  2. [Section 3.1] The statement that exact triples (f,σ,v) 'are typically unique to individual trees and are not shared or reused across trees' is imprecise: split values drawn from the same feature range can coincide across trees, and the argument for aggregation does not require uniqueness.
  3. [Section 3.1, Eq. (2)] The IOP-Score definition refers to 'the frequency of the edge connecting node v to the Inlier class' and 'to the Outlier class', but the graph construction described earlier creates edges between predicates, not between predicates and terminal class nodes. Please clarify how class nodes are incorporated into the graph and whether the frequencies in Eq. (2) are the weighted frequencies introduced in Eq. (1) or the raw frequencies.
  4. [Algorithm 1] Line 3 extracts predicates as triples (f,σ,v), but line 11 converts them to pairs (f,σ) only after the graph has already been described as having predicate pairs as nodes; a reader cannot tell when the value removal happens relative to graph construction.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the IOP-Score is a defined summary of iForest traversals, not a fitted prediction; the only self-reference is a non-load-bearing background citation to the authors' earlier DPG work.

full rationale

The paper's construction chain is transparent: predicates are extracted directly from iForest split rules (Section 3.1), each training sample's traversal generates predicate lists, edges are the weighted frequencies of consecutive predicates, and the IOP-Score (Eq. 2) is a normalized class-conditional frequency. No parameter is fitted to a subset and then renamed a prediction; Eq. (1) weights are fixed inverse class counts, and Eq. (2) is a definition, so saying a negative IOP-Score indicates outlier-associated predicates is restating the definition rather than deriving an independent result. The synthetic experiments are in-sample sanity checks of whether the graph recovers the features used to generate outliers, which is appropriate for a post-hoc explanation method and does not reduce to a prediction. The deletion of split values and the max-depth filtering are explicit design choices that may limit fidelity (the graph cannot represent exact thresholds, and long-path outlier traversals are discarded), but these are correctness limitations, not circular reasoning. The only self-citation is to the earlier DPG paper [12], from which the graph-construction framework is taken; that paper is cited as background and the present method's adaptation to iForest, value abstraction, weighting, and IOP-Score are new content, so the self-citation is not load-bearing.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The method introduces no fitted parameters; the weights and scores are computed deterministically from the model and class labels. The main added assumptions are design choices about which parts of the iForest structure to keep (directions, but not values) and which to discard (deep outlier paths). These are load-bearing for the explanation's faithfulness.

assumptions (4)
  • ad hoc to paper Predicates stripped of their split values (feature, direction only) preserve the information needed to explain iForest's behavior.
    The method is built entirely on these abstracted predicates; no experiment verifies that threshold-free aggregation preserves model logic.
  • ad hoc to paper Predicate lists longer than the tree's maximum depth can be discarded for outliers without distorting the explanation.
    Section 3.1 removes these paths to avoid "ambiguity", but this is an unvalidated post-hoc exclusion that could hide evidence about outlier behavior.
  • domain assumption Inverse-frequency edge weighting makes inlier and outlier transition counts comparable.
    Eq. (1) reweights transitions by class counts; this is a plausible heuristic but there is no justification that it yields a faithful class balance.
  • domain assumption Scikit-learn's Isolation Forest implementation faithfully realizes the original iForest algorithm.
    All experiments rely on the sklearn implementation, including its max-depth and scoring behavior, which may differ from the original 2008 formulation in details.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Extending Decision Predicate Graphs for Comprehensive Explanation of Isolation Forest." pith.science (2026). https://pith.science/paper/R7JYOYF5

@misc{pith2026250504019,
  author       = {Pith},
  title        = {Pith review of: Extending Decision Predicate Graphs for Comprehensive Explanation of Isolation Forest},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R7JYOYF5}},
  note         = {Machine review of arXiv:2505.04019}
}
read the original abstract

The need to explain predictive models is well-established in modern machine learning. However, beyond model interpretability, understanding pre-processing methods is equally essential. Understanding how data modifications impact model performance improvements and potential biases and promoting a reliable pipeline is mandatory for developing robust machine learning solutions. Isolation Forest (iForest) is a widely used technique for outlier detection that performs well. Its effectiveness increases with the number of tree-based learners. However, this also complicates the explanation of outlier selection and the decision boundaries for inliers. This research introduces a novel Explainable AI (XAI) method, tackling the problem of global explainability. In detail, it aims to offer a global explanation for outlier detection to address its opaque nature. Our approach is based on the Decision Predicate Graph (DPG), which clarifies the logic of ensemble methods and provides both insights and a graph-based metric to explain how samples are identified as outliers using the proposed Inlier-Outlier Propagation Score (IOP-Score). Our proposal enhances iForest's explainability and provides a comprehensive view of the decision-making process, detailing which features contribute to outlier identification and how the model utilizes them. This method advances the state-of-the-art by providing insights into decision boundaries and a comprehensive view of holistic feature usage in outlier identification. -- thus promoting a fully explainable machine learning pipeline.

Figures

Figures reproduced from arXiv: 2505.04019 by the authors.

Figure 1
Figure 1. Schematic demonstration of how DPG works. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed approach: iForest DPG representation. Predicates are represented as triples ( [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Pairplot of the first synthetic dataset. The dataset comprises [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Global representation of the iForest model as a DPG produced by our method for the first synthetic dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Pairplot of the second synthetic dataset. The dataset comprises [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Global representation of the iForest model as a DPG produced by our method for the second synthetic dataset. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Global representation of the iForest model as a DPG structure produced by our method for the Annthyroid [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 12 canonical work pages

  1. [1]

    Zachary C. Lipton. The mythos of model interpretability: In machine learning, the concept of interpretability is both important and slippery. Queue, 16(3):31–57, 2018. ISSN 1542-7730, 1542-7749. doi:10.1145/3236386.3241340

    arXiv 2018

  2. [2]

    Measuring the quality of explanations: The system causability scale (scs)

    Andreas Holzinger, Amanda Carrington, and Heiko Muller. Measuring the quality of explanations: The system causability scale (scs). KI - Künstliche Intelligenz, 34(2):193–198, 2020. ISSN 1610-1987. doi:10.1007/s13218- 020-00636-z

    work page doi:10.1007/s13218- 2020

  3. [3]

    Development of a human-centred psychometric test for the evaluation of ex- planations produced by XAI methods

    Giulia Vilone and Luca Longo. Development of a human-centred psychometric test for the evaluation of ex- planations produced by XAI methods. In Explainable Artificial Intelligence, pages 205–232. Springer Nature Switzerland, 2023. ISBN 978-3-031-44070-0. doi:10.1007/978-3-031-44070-0_11

    work page doi:10.1007/978-3-031-44070-0_11 2023

  4. [4]

    A general-purpose method for applying explainable AI for anomaly detection

    John Sipple and Abdou Youssef. A general-purpose method for applying explainable AI for anomaly detection. In International Symposium on Methodologies for Intelligent Systems, pages 162–174. Springer, 2022. ISBN 978-3-031-16564-1. doi:10.1007/978-3-031-16564-1_16

    work page doi:10.1007/978-3-031-16564-1_16 2022

  5. [5]

    Transparent data preprocessing for machine learning

    Sebastian Strasser and Meike Klettke. Transparent data preprocessing for machine learning. In Proceedings of the 2024 Workshop on Human-In-the-Loop Data Analytics, pages 1–6, New York, NY , USA, 2024. Association for Computing Machinery. ISBN 9798400706936. doi:10.1145/3665939.3665960

    arXiv 2024

  6. [6]

    Towards explaining the effects of data preprocessing on machine learning

    Carlos Vladimiro González Zelaya. Towards explaining the effects of data preprocessing on machine learning. In 2019 IEEE 35th International Conference on Data Engineering (ICDE) , pages 2086–2090. IEEE, 2019. doi:10.1109/ICDE.2019.00245

    arXiv 2019

  7. [7]

    Isolation forest

    Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. Isolation forest. In 2008 Eighth IEEE International Conference on Data Mining, pages 413–422. IEEE, 2008. doi:10.1109/ICDM.2008.17

    doi:10.1109/icdm.2008.17 2008

  8. [8]

    Why Are You Weird? Infusing Interpretability in Isolation Forest for Anomaly Detection

    Nirmal Sobha Kartha, Clement Gautrais, and Vincent Vercruyssen. Why are you weird? infusing interpretability in isolation forest for anomaly detection. arXiv preprint arXiv:2112.06858, 2021. doi:10.48550/arXiv.2112.06858

    work page Pith review arXiv doi:10.48550/arxiv.2112.06858 2021

Show all 23 references
  1. [9]

    Enhancing explainability in predictive maintenance: Investigating the impact of data preprocessing techniques on xai effectiveness

    Mouhamadou-Lamine Ndao, Genane Youness, Ndèye Niang, and Gilbert Saporta. Enhancing explainability in predictive maintenance: Investigating the impact of data preprocessing techniques on xai effectiveness. In The 37th International Conference of the Florida Artificial Intellig...

    2024 doi

  2. [10]

    Lundberg and Su-In Lee

    Scott M. Lundberg and Su-In Lee. A unified approach to interpreting model predictions. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 4768–4777, 2017. ISBN 9781510860964

    2017

  3. [11]

    Interpretable anomaly detection with diffi: Depth-based feature importance of isolation forest

    Mattia Carletti, Matteo Terzi, and Gian Antonio Susto. Interpretable anomaly detection with diffi: Depth-based feature importance of isolation forest. Engineering Applications of Artificial Intelligence , 119:105730, 2023. ISSN 0952-1976. doi:10.1016/j.engappai.2022.105730. 15...

    2023

  4. [12]

    Decision predicate graphs: Enhancing interpretability in tree ensembles

    Leonardo Arrighi, Luca Pennella, Gabriel Marques Tavares, and Sylvio Barbon Junior. Decision predicate graphs: Enhancing interpretability in tree ensembles. In World Conference on Explainable Artificial Intelligence, pages 311–332. Springer Nature Switzerland, 2024. ISBN 978-3...

    2024 doi

  5. [13]

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

    Timo Speith. A review of taxonomies of explainable artificial intelligence (XAI) methods. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’22, pages 2239–2250. Association for Computing Machinery, 2022. ISBN 978-1-4503-9352-2. doi...

    2022

  6. [14]

    A qualitative investigation of the explainability of defeasible argumentation and non-monotonic fuzzy reasoning

    Lucas Rizzo and Luca Longo. A qualitative investigation of the explainability of defeasible argumentation and non-monotonic fuzzy reasoning. In Proceedings for the 26th AIAI Irish Conference on Artificial Intelligence and Cognitive Science Trinity College Dublin, Dublin, Irela...

    2018 doi

  7. [15]

    Enhancing explainability in mobility data science through a combination of methods

    Georgios Makridis, Vasileios Koukos, Georgios Fatouros, Maria Margarita Separdani, and Dimosthenis Kyriazis. Enhancing explainability in mobility data science through a combination of methods. In Intelligent Computing, pages 45–60. Springer, 2024. ISBN 978-3-031-62269-4. doi:1...

    2024 doi

  8. [16]

    Isolation forest with exclusion of attributes based on shapley index

    Lukasz Rachwał, Bartosz Krawczyk, and Michał Wo´ zniak. Isolation forest with exclusion of attributes based on shapley index. IEEE Transactions on Neural Networks and Learning Systems , 34(8):4011–4022, 2023. doi:10.1109/ACCESS.2024.3432174

    2023

  9. [17]

    Anomaly detection and explanation in coal data using isolation forest, random forest, and shap

    Yang Liu and Chris Aldrich. Anomaly detection and explanation in coal data using isolation forest, random forest, and shap. International Journal of Coal Geology, 250:103921, 2023. doi:10.1016/j.coal.2023.103921

    2023

  10. [18]

    Enhancing interpretability and generalizability in extended isolation forests

    Alessio Arcudi, Davide Frizzo, Chiara Masiero, and Gian Antonio Susto. Enhancing interpretability and generalizability in extended isolation forests. Engineering Applications of Artificial Intelligence, 138:109409,

  11. [19]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python. Journal of Machine Learni...

    2011

  12. [20]

    Unsupervised Anomaly Detection Benchmark, 2015

    Markus Goldstein. Unsupervised Anomaly Detection Benchmark, 2015

    2015

  13. [21]

    A comparative evaluation of unsupervised anomaly detection algorithms for multivariate data

    Markus Goldstein and Seiichi Uchida. A comparative evaluation of unsupervised anomaly detection algorithms for multivariate data. PLoS ONE, 11(4):e0152173, 2016. doi:10.1371/journal.pone.0152173

    2016 doi

  14. [22]

    Bache and M

    K. Bache and M. Lichman. UCI machine learning repository. URL http://archive.ics.uci.edu/ml. 16

  15. [2024]

    doi:10.1016/j.engappai.2024.109409

    ISSN 0952-1976. doi:10.1016/j.engappai.2024.109409

    1976

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.