LatentGraph: from latent states to rule-based expressions for explainable financial forecasting
LatentGraph: from latent states to rule-based expressions for explainable financial forecasting
Financial risk management needs forecasting tools that explain their reasoning, since accuracy alone is insufficient for board decisions. Accordingly, the advancement of classification models demands end-to-end transparency for the decisions they make. To address this, this paper presents LatentGraph, which aims for explainability first while keeping competitive accuracy. The pipeline encodes multivariate data into latent factors that pass through a lightweight feature graph, then a learnable binarization stage, and finally a Tsetlin Machine that writes short if-then rules. Every stage logs numeric summaries, thresholds, and figures that allow each decision to trace back to concrete coordinates and profiles without extra tooling. We assess LatentGraph on the public Telco Customer Churn dataset. The untuned LatentGraph attains the highest recall at 0.7096 among compared models, which remains close to classical baselines, while also providing clause level explanations with per-clause support and confidence.
Binarization, Explainable AI, Graph Embedding, Latents, Time Series, Time Shape Encodings, Transparency, Tsetlin Machines
745-752
Association for Computing Machinery
Erkus, Ekin Can
81be5b02-a82f-415d-957f-c5ca280684d7
Chan, Alex
4ce24eb6-160b-4c58-94eb-a4ecd5586166
Distaso, Walter
333a4736-8f1a-44b8-b6db-ed1528bed743
Thomas, David
5701997d-7de3-4e57-a802-ea2bd3e6ab6c
Yakovlev, Alex
d6c94911-c126-4cb7-8f92-d71a898ebbb2
Shafik, Rishad
aa0bdafc-b022-4cb2-a8ef-4bf8a03ba524
14 November 2025
Erkus, Ekin Can
81be5b02-a82f-415d-957f-c5ca280684d7
Chan, Alex
4ce24eb6-160b-4c58-94eb-a4ecd5586166
Distaso, Walter
333a4736-8f1a-44b8-b6db-ed1528bed743
Thomas, David
5701997d-7de3-4e57-a802-ea2bd3e6ab6c
Yakovlev, Alex
d6c94911-c126-4cb7-8f92-d71a898ebbb2
Shafik, Rishad
aa0bdafc-b022-4cb2-a8ef-4bf8a03ba524
Erkus, Ekin Can, Chan, Alex, Distaso, Walter, Thomas, David, Yakovlev, Alex and Shafik, Rishad
(2025)
LatentGraph: from latent states to rule-based expressions for explainable financial forecasting.
In ICAIF 2025 - 6th ACM International Conference on AI in Finance.
Association for Computing Machinery.
.
(doi:10.1145/3768292.3770428).
Record type:
Conference or Workshop Item
(Paper)
Abstract
Financial risk management needs forecasting tools that explain their reasoning, since accuracy alone is insufficient for board decisions. Accordingly, the advancement of classification models demands end-to-end transparency for the decisions they make. To address this, this paper presents LatentGraph, which aims for explainability first while keeping competitive accuracy. The pipeline encodes multivariate data into latent factors that pass through a lightweight feature graph, then a learnable binarization stage, and finally a Tsetlin Machine that writes short if-then rules. Every stage logs numeric summaries, thresholds, and figures that allow each decision to trace back to concrete coordinates and profiles without extra tooling. We assess LatentGraph on the public Telco Customer Churn dataset. The untuned LatentGraph attains the highest recall at 0.7096 among compared models, which remains close to classical baselines, while also providing clause level explanations with per-clause support and confidence.
Text
3768292.3770428
- Version of Record
More information
Published date: 14 November 2025
Venue - Dates:
6th ACM International Conference on AI in Finance, ICAIF 2025, , Singapore, Singapore, 2025-11-15 - 2025-11-18
Keywords:
Binarization, Explainable AI, Graph Embedding, Latents, Time Series, Time Shape Encodings, Transparency, Tsetlin Machines
Identifiers
Local EPrints ID: 512938
URI: https://eprints.soton.ac.uk/id/eprint/512938
PURE UUID: 0c139a1f-206a-461b-99d9-caef8aae01a2
Catalogue record
Date deposited: 28 Jul 2026 16:45
Last modified: 13 Sep 2026 15:49
Export record
Altmetrics
Contributors
Author:
Ekin Can Erkus
Author:
Alex Chan
Author:
Walter Distaso
Author:
David Thomas
Author:
Alex Yakovlev
Author:
Rishad Shafik
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics