X hits on this document

PDF document

MODELING HUMAN FACTORS INVOLVED IN CHEMICAL/BIOLOGICAL WARNING & REPORTING - page 2 / 6

21 views

0 shares

0 downloads

0 comments

2 / 6

BACKGROUND

Methods for conducting CTA have evolved from the study of Naturalistic Decision Making (NDM), a field of study pioneered at Klein Associates Division of ARA. Naturalistic Decision Making is the study of how people perform and make decisions under conditions of stress, time pressure, high consequences, and ambiguity1. A major objective of NDM is to understand the cognitive processes employed during task performance within the operational environment.

CTA comprises a series of techniques for knowledge elicitation and knowledge representation2. We have found that no single method works well in all cases  they must be combined and adapted to suit the needs of each domain. The methods that are most effective in each domain depend on the characteristics of the task, the characteristics of the operators, and the conditions under which they must perform the task. The CTA toolkit includes a number of different knowledge elicitation techniques that have evolved over the past two decades. For this project, our primary CTA have included a combination of incident-based techniques for both individuals and teams. These methods have been combined with naturalistic observation of the operator within his operational environment.

The HFA is based on a Bayesian Recognition Decision Model (BRDM)3, a Bayesian implementation of the recognition-primed decision model4 based primarily on models of episodic recognition memory5,6. Bayesian models of psychological processes offer principled approaches to understanding how evidence is used in various decision processes. These include decisions about episodic memory5,7, semantic knowledge6, perceptual judgment8, and word recognition9.

The challenges faced by the HFA include a low-incidence of actual CBRN events combined with a high-incidence of false alarms and incomplete data. The HFA must de able to discern real evidence of a CRBN attack from false alarms that could come from faulty sensors, panicked individuals, or a lack of pertinent data. The BRDM model fits the requirements of the HFA by accounting for sensor reliability, the likelihoods of certain CBRN events taking place (based on daily situational reports), and the likelihood that observed evidence of a CBRN attack can be present in the absence of a CBRN attack. In order to create the HFA, the BRDM is paired with a procedural component that looks for information to prove or falsify the existence of a CBRN attack via available resources and sensors.

The underlying BRDM equation that determines if datum i is evidence towards CBRN event k for the HFA is given by:

where:

ik denotes the likelihood ratio that evidence i pertains to CBRN event k, r is the HFA estimate of a sensors reliability

G ik is the HFA estimate of the binomial probability of evidence i for CBRN event k,

  • ik

 rG ik (1rF i )

Fi

(1)

  • F i is the HFA estimate of the likelihood that observed evidence i can be present in the

absence of a CBRN attack.

The likelihood ratio of CBRN event k given all available evidence is given by:

2

Document info
Document views21
Page views21
Page last viewedThu Dec 08 02:18:44 UTC 2016
Pages6
Paragraphs88
Words1519

Comments