Figure 4
From: Algorithmic discovery of dynamic models from infectious disease data
Comparison between chickenpox incidence data and coefficients of the best SINDy-discovered model using a function library of polynomials up to 2nd order, and showing the model with the lowest AIC scores across the \({S}_{0}-\lambda \) parameter grid. The discovered model accurately replicates the annual cycle present in the data in both the susceptible and infection classes. As in the measles case, it also identifies a strong dependence on the mass action incidence term in both the \(S\) and \(i\) equations. Note also that the coefficient of \(S\) and \(I\) in their respective equations are close to 1, as expected in discrete disease models. The sparse regression excluded six terms, giving \(r=0.25\). The parameter grids appear in SI Appendix, Fig. 7. The results in the table display the SINDy-discovered coefficients of the corresponding terms in (Eqs. 1–3).
