EURO 2025 Leeds
Abstract Submission

1787. Optimizing Professional Judgement in Insurance Risk Assessment: A Hybrid AI Approach Using Evidential Reasoning and Belief Rule Base

Invited abstract in session TB-28: Multi-Agent Systems and Reinforcement Learning for Decision Support, stream Decision Support Systems.

Tuesday, 10:30-12:00
Room: Maurice Keyworth 1.03

Authors (first author is the speaker)

1. Karim Derrick
Alliance Business School, University of Manchester
2. Xi Liu
Kennedy Law LLP
3. Joe Cunningham
Kennedys IQ
4. Jian-Bo Yang
Alliance Manchester Business School, The University of Manchester
5. Dong-Ling Xu
Alliance Manchester Business School, The University of Manchester

Abstract

Professional judgement in insurance, legal, and financial risk assessment is inconsistent and subject to cognitive biases, yet remains central to decision-making. Large Language Models (LLMs) offer new opportunities but are probabilistic, prone to hallucination, and lack explainability. This paper introduces SmartRisk, a hybrid AI framework combining LLMs for attribute extraction with a Belief Rule Base (BRB) model and an evidential reasoning engine to improve accuracy, consistency, and explainability in risk assessment.

Grounded in Operations Research (OR), Multi-Attribute Decision Making (MADM), and Evidential Reasoning (ER), we develop a neuro-symbolic decision model integrating expert knowledge with machine learning. Our approach captures the nonlinear nature of risk assessment, supporting uncertainty representation, dynamic evidence updating, and expert-guided optimization.

Empirical trials demonstrate that SmartRisk:

-Eliminates inter-rater inconsistency (expert assessments α < 0.4).
-Enhances speed—reducing policy and claims analysis from hours to minutes.
-Mitigates AI biases—constraining LLMs to structured extraction and explainable inference.

This research extends OR methodologies into decision automation, showing that hybrid AI frameworks leveraging BRB and LLMs will define the future of intelligent risk assessment—balancing automation, accountability, and optimal decision support.

Keywords

Status: accepted

Back to the list of papers