[Paper Review] Threat assessment of a possible Vehicle-Born Improvised Explosive Device using DSmT

Jean Dezert, Florentín Smarandache|arXiv (Cornell University)|Aug 2, 2010

Fire Detection and Safety Systems9 references3 citations

TL;DR

This paper applies Dezert-Smarandache Theory (DSmT) to assess the threat of a Vehicle-Born Improvised Explosive Device (VBIED), fusing uncertain, imprecise, and conflicting evidence from analysts and automated systems. It demonstrates that using refined qualitative approximation with PCR5 and PCR6 rules leads to a consistent decision to evacuate the building, even when sources conflict or vary in reliability.

ABSTRACT

This paper presents the solution about the threat of a VBIED (Vehicle-Born Improvised Explosive Device) obtained with the DSmT (Dezert-Smarandache Theory). This problem has been proposed recently to the authors by Simon Maskell and John Lavery as a typical illustrative example to try to compare the different approaches for dealing with uncertainty for decision-making support. The purpose of this paper is to show in details how a solid justified solution can be obtained from DSmT approach and its fusion rules thanks to a proper modeling of the belief functions involved in this problem.

Motivation & Objective

• To address the challenge of threat assessment in VBIED scenarios with incomplete, imprecise, and conflicting information.
• To evaluate whether human expertise (analyst) or automated systems (ANPR) provide more reliable input for decision-making.
• To model and quantify uncertainty in belief functions using DSmT, particularly under conditions of partial ignorance and conflicting evidence.
• To compare the performance of different fusion rules (PCR5 and PCR6) in supporting security decisions under uncertainty.
• To provide a justified, quantitative decision-support framework for evacuation decisions in high-stakes security scenarios.

Proposed method

• Model the VBIED problem using three marginal frames: individual identity (A/¬A), vehicle type (V/¬V), and proximity to building (B/¬B), each defined on their respective power sets.
• Construct a joint frame Θ = Θ₁ × Θ₂ × Θ₃ with 8 possible states representing combinations of identity, vehicle, and location.
• Assign basic belief assignments (bba) to each source based on observations: Analyst 1 (high experience), ANPR (30% probability), and Analyst 2 (inexperienced).
• Apply qualitative DSmT fusion rules (qPCR5 and qPCR6) using refined label approximations to normalize qualitative belief masses and handle imprecision.
• Use the generalized pignistic transformation (qDSmP) to compute final decision-support probabilities for key hypotheses.
• Evaluate decision outcomes using belief (Bel), plausibility (Pl), and qDSmP values across different decision-support hypotheses (e.g., θ₈, θ₇∪θ₈, θ₆∪θ₇∪θ₈).

Experimental results

Research questions

• RQ1 What is the optimal fusion strategy for combining expert judgment and automated system outputs in VBIED threat assessment?
• RQ2 How does the reliability of an experienced analyst compare to that of an automated ANPR system in a high-uncertainty context?
• RQ3 Can qualitative DSmT fusion rules (PCR5 and PCR6) produce consistent and justifiable evacuation decisions under conflicting and imprecise evidence?
• RQ4 How does the use of refined label approximation affect the normalization and interpretability of qualitative belief masses?
• RQ5 Which decision-support hypothesis (e.g., θ₈, θ₇∪θ₈, or θ₆∪θ₇∪θ₈) leads to the most prudent and consistent evacuation decision?

Key findings

• Using refined label approximation in qDSmP ensures normalized qualitative probabilities, unlike crude approximations which yield non-normalized results.
• For the most prudent decision-support hypothesis (θ₆ ∪ θ₇ ∪ θ₈), both PCR5 and PCR6 rules yield a belief interval [L₁.₂₈, L₄] and plausibility [L₀, L₂.₇₂], indicating strong support for evacuation.
• The belief in θ₈ (A is near B in a white Toyota) reaches [L₀.₈₁, L₃.₁₂] under PCR5 and [L₀.₈₁, L₃.₂₄] under PCR6, indicating high confidence in the threat.
• PCR5 suggests not evacuating when using θ₈ as the sole hypothesis due to lower belief, while PCR6 recommends evacuation—highlighting the sensitivity of results to fusion rule choice.
• Despite differences between PCR5 and PCR6, both rules agree on evacuation when using the comprehensive hypothesis θ₆ ∪ θ₇ ∪ θ₈, which captures all dangerous configurations.
• The analysis confirms that prior information and source reliability significantly influence final decisions, and DSmT provides a robust framework for integrating such heterogeneous inputs.