1: \begin{abstract}

2: Prediction of dynamical time series with additive noise using support

3: vector machines or kernel based regression has been proved to be consistent

4: for certain classes of discrete dynamical systems. Consistency implies

5: that these methods are effective at computing the expected value of

6: a point at a future time given the present coordinates. However, the

7: present coordinates themselves are noisy, and therefore, these methods

8: are not necessarily effective at removing noise. In this article,

9: we consider denoising and prediction as separate problems for flows,

10: as opposed to discrete time dynamical systems, and show that the use

11: of smooth splines is more effective at removing noise. Combination

12: of smooth splines and kernel based regression yields predictors that

13: are more accurate on benchmarks typically by a factor of 2 or more.

14: We prove that kernel based regression in combination with smooth splines

15: converges to the exact predictor for time series extracted from any

16: compact invariant set of any sufficiently smooth flow. As a consequence

17: of convergence, one can find examples where the combination of kernel

18: based regression with smooth splines is superior by even a factor

19: of $100$. The predictors that we compute operate on delay coordinate

20: data and not the full state vector, which is typically not observable.

21: \end{abstract}

22: