Detecting nonlinearity and chaos in epidemic data