One of the most crucial yet poorly constrained parameters in modelling the ionizing emissivity is the escape fraction of photons from star-forming galaxies. Several theoretical and observational studies have been conducted over the past few years, but consensus regarding its redshift evolution has yet to be achieved. We present here the first non-parametric reconstruction of this parameter as a function of redshift from a data-driven reionization model using a Gaussian Process Regression method. Our finding suggests a mild redshift evolution of escape fraction with a mean value of 4%,7%,∼10% at z=2,6,12. However, a constant escape fraction of 6−10% at z≳6 is still allowed by current data and also matches other reionization-related observations. With the detection of fainter high redshift galaxies from upcoming observations of JWST, the approach presented here will be a robust tool to put the most stringent constraint on escape fraction as well as reionization histories.