The success of ground-based instruments for high contrast exoplanet imaging depends on the degree to which adaptive optics (AO) systems can mitigate atmospheric turbulence. While modern AO systems typically suffer from millisecond time lags between wavefront measurement and control, predictive wavefront control (pWFC) is a means of compensating for those time lags using previous wavefront measurements, thereby improving the raw contrast in the post-coronagraphic science focal plane. A method of predictive control based on Empirical Orthogonal Functions (EOF) has previously been proposed and demonstrated on Subaru/SCExAO. In this paper we present initial tests of this method for application to the near-infrared pyramid wavefront sensor (PYWFS) recently installed in the Keck II AO system. We demonstrate the expected root-mean-square (RMS) wavefront error and contrast benefits of pWFC based on simulations, applying pWFC to on-sky telemetry data saved during commissioning of the PYWFS. We discuss how the performance varies as different temporal and spatial scales are included in the computation of the predictive filter. We further describe the implementation of EOF pWFC within the PYWFS dedicated real-time controller (RTC), and, via daytime testing at the observatory, we demonstrate the performance of pWFC in real time when pre-computed phase screens are applied to the deformable mirror (DM).