Data-driven prediction methods are getting popular as the availability of quality data is increasing. One of the features of these methods is that they can be used to circumvent the non-linearities of dynamical systems. This is done via the Koopman operator. In this talk, I will give an overview of "Kernel Analog Forecasting" which uses concepts from operator theory, kernel methods and machine learning. If time permits, I will give a small example towards the end.