In order to estimate the threshold of photoelectron yield spectroscopy (PYS), regression estimator was created by machine learning using calculation data based on Fowler's theory. We compared verification data with the prediction by regression estimators. The verification data measured by AC-3 instrument manufactured by Riken Keiki Co. are the number of 87 (Au : 31, Metals except for Au : 16, oxide · semiconductor : 15, organic : 25) with estimate value by analyst (label data). As a result, the prediction performance was 70% over in the range of difference between label data and estimation from ±0.2 eV. On the other hand, spectra in the difference of 0.3 eV were included in difficult labeling by analysts. The prediction by machine learning gives the guide to analysis.