In classical design theory one traditionally proposes experiments involving only 2-level factors. For these two level designs the theory for reduced (fractional) factorial 2x-z was developed and refined during the late 1950s. There are many arguments that encourage this but there exists however a market demand for algorithms that also handle the existence of multilevel (ML) factors, i.e. factors with more than 2-levels, in an experiment, and in a similar way make reduced designs for them.
In theory the existence of perfect solutions to such problems are strictly limited to the set of orthogonal arrays. In practice, we can however do with designs that are not perfectly but nearly orthogonal. In the latter case, we need to be able to find these solutions.
Subject for thesis project:
Literature study: What has already been done on this subject and which algorithms and evaluating methods are currently in market use.
Create and evaluate different criteria for quality-measure of ML-designs such as balance, efficiency, orthogonality and max-min distance.
Investigate possibilities to form efficient designs with evolutionary algorithms, possibly combined with regular D-optimal algorithms.
Projektet är tänkt för TM-programmet
Speciella förkunskapskrav FFR105 Stochastic optimization algorithms, MSG800 Basic stochastic processes, MVE095 Optioner och matematik, samt minst 37.5 hp godkända i nationalekonomi
Handledare Mattias Sunden
Examinator Maria Roginskaya
Institution Matematiska vetenskaper