%0 Journal Article %1 Holme2015Time %A Holme, Petter %A Takaguchi, Taro %D 2015 %J Physical Review E %K social-networks predictive-models epidemic-models epidemics %N 4 %R 10.1103/PhysRevE.91.042811 %T Time evolution of predictability of epidemics on networks %U http://dx.doi.org/10.1103/PhysRevE.91.042811 %V 91 %X Epidemic outbreaks of new pathogens, or known pathogens in new populations, spread fear much because they are hard to predict. For theoretical models of disease spreading, on the other hand, quantities characterizing the outbreak converge to deterministic functions of time. Our goal in this paper is to shed some light on this apparent discrepancy. We measure the diversity of (and, thus, the predictability of) outbreak sizes and extinction times as functions of time given different scenarios of the amount of information available. Under the assumption of perfect information -- i.e. knowing the state of each individual with respect to the disease -- the predictability decreases exponentially, or faster, with time. The decay is slowest for intermediate values of the per-contact transmission probability. With a weaker assumption on the information available, assuming that we know only the fraction of currently infectious, recovered or susceptible individuals, the predictability also decreases exponentially most of the time. There are, however, some peculiar regions in this scenario where the predictability decreases. In other words, to predict its final size with a given accuracy, we would need increasingly much information about the outbreak.

@article{Holme2015Time, abstract = {{Epidemic outbreaks of new pathogens, or known pathogens in new populations, spread fear much because they are hard to predict. For theoretical models of disease spreading, on the other hand, quantities characterizing the outbreak converge to deterministic functions of time. Our goal in this paper is to shed some light on this apparent discrepancy. We measure the diversity of (and, thus, the predictability of) outbreak sizes and extinction times as functions of time given different scenarios of the amount of information available. Under the assumption of perfect information -- i.e. knowing the state of each individual with respect to the disease -- the predictability decreases exponentially, or faster, with time. The decay is slowest for intermediate values of the per-contact transmission probability. With a weaker assumption on the information available, assuming that we know only the fraction of currently infectious, recovered or susceptible individuals, the predictability also decreases exponentially most of the time. There are, however, some peculiar regions in this scenario where the predictability decreases. In other words, to predict its final size with a given accuracy, we would need increasingly much information about the outbreak.}}, added-at = {2019-06-10T14:53:09.000+0200}, archiveprefix = {arXiv}, author = {Holme, Petter and Takaguchi, Taro}, biburl = {https://www.bibsonomy.org/bibtex/2b1ff1bba9163e0417772a81e2a3a18de/nonancourt}, citeulike-article-id = {13463097}, citeulike-linkout-0 = {http://dx.doi.org/10.1103/PhysRevE.91.042811}, citeulike-linkout-1 = {http://arxiv.org/abs/1412.4875}, citeulike-linkout-2 = {http://arxiv.org/pdf/1412.4875}, day = 28, doi = {10.1103/PhysRevE.91.042811}, eprint = {1412.4875}, interhash = {3e49f702b33d184efa7feb39b36fa21e}, intrahash = {b1ff1bba9163e0417772a81e2a3a18de}, issn = {1550-2376}, journal = {Physical Review E}, keywords = {social-networks predictive-models epidemic-models epidemics}, month = apr, number = 4, posted-at = {2014-12-17 15:16:05}, priority = {2}, timestamp = {2019-07-31T12:35:19.000+0200}, title = {{Time evolution of predictability of epidemics on networks}}, url = {http://dx.doi.org/10.1103/PhysRevE.91.042811}, volume = 91, year = 2015 }