Parameter selection for modeling of epidemic networks

Abstract

The accurate modeling of epidemics on social contact networks is difficult due to the variation between different epidemics and the large number of parameters inherent to the problem. To reduce complexity, evolutionary computation is used to create a generative representation of the epidemic model. Previous gains from the use of local, verses global, operators are further explored to better balance exploration and exploitation of the genetic algorithm. A typical parameter study is conducted to test this new local operator and the new method of point packing is utilized as a proof of concept to perform a better search of the parameter space. All experiments from both approaches are tested against nine epidemic profiles. The point-packing driven parameter search demonstrates both that the algorithm parameters interact substantially and in a non-linear fashion and also shows that the good parameter settings are problem specific.

Representation for Evolution of Epidemic Models

“…A caveat to these findings is that these are slight trends in that the confidence intervals of all PSs within a given profile overlap. These findings correlate with those found in [11] where local toggle performed better on all profiles except profile 8. This provides further evidence that better performance may be achieved if parameters are tuned on a per-profile basis.…”

“…In [17] it is demonstrated that swap allowed for better performance than hop for the PM problem. Additionally, [11] establishes that, for most profiles, the best fitness was accomplished when the probability of toggle and local toggle are approximately equal. Therefore, the percentage for swap is zero for both parameter sweeps.…”

“…Edge swap was used because it was able to take an initial personal contact network and change which members of the population come into contact with each other. Additional operations were added to the representation in [1], allowing for the addition and deletion of connections; more recent work has focused on adding local variants of the operations [8], [10], [11]. Local operations are only applied to triples of vertices for which any pair of the vertices are at a maximum distance of two before and after the operation is applied.…”

“…PSs 20, 23, and 27 yield good results across the four environments used for testing. PS 20 consisted of mainly add and local add; it is common for a point packing to include an PS exhibiting these characteristics to be among the best PSs, regardless of problem, see [11]. Both PSs 23 and 27 have large amounts of null probability with PS 27 also having a significant portion attributed to local add.…”

Modelling of Vaccination Strategies for Epidemics using Evolutionary Computation

“…The Local THADS-N generative representation [11] will be used in this paper to generate the personal contact networks which then are assessed for performance against the test problems. Existing applications of generative representations include [9], [13], [17].…”