Capturing the impact of critical criteria in Multiple-Criteria Decision Analysis

Add to your list(s) Download to your calendar using vCal

• John Moore (QINETIQ)
• Wednesday 20 February 2019, 12:00-13:00
• Room R17/18, Watson building.

If you have a question about this talk, please contact Sergey Sergeev.

The Multiple-Criteria Decision Analysis (MCDA) discipline seeks to provide rational and auditable methods of evaluating the options available to an enterprise against a comprehensive range of decision criteria, even when these involve conflicting priorities. Current practice in MCDA is dominated by linear methods which, in effect, score each competing option or offering as a sum of its scores against a number of decision criteria, weighted according to their relative importance. These methods offer poor fidelity when there are numerous criteria that are considered “critical”, in that failing to achieve an acceptable outcome would negate any level of success against the other criteria. This is because the difference between success and failure against any single criterion cannot exceed the weight assigned to that criterion, relative to the sum of criterion weights. In a major analysis, possibly involving hundreds of criteria, the proportion of criteria that can be assigned genuinely high “criticality” is therefore drastically limited: for example, no more than four can each contribute >20% of the total score. In 2013, I proposed a “multiplicative” objective function, parameterised directly by the criticality assigned to each criterion, in which such assignments were unconstrained and their significance captured in full. Subsequent experimentation showed a need to refine the function, to avoid anomalous behaviours when the function was applied to decisions in which the criteria are structured in a hierarchy. The definitive function automatically generates a set of auxiliary parameters, provisionally termed “q-values”, which stabilise the response properties in these problematic cases. The definitive function allows judgements of criterion criticality to be represented in MCDA problems with far greater fidelity than is offered by linear techniques, at any level of complexity, and over any realistic range of possible offerings.

This talk is part of the Optimisation and Numerical Analysis Seminars series.

This talk is included in these lists:

• Optimisation and Numerical Analysis Seminars
• Room R17/18, Watson building
• School of Mathematics Events

Note that ex-directory lists are not shown.