Investigation of rigidity dependence of long-term cosmic ray variation with cosmic ray muon count rate on the ground requires correction for atmospheric pressure and temperature effect. Although a correction method developed by Mendoça [1] works well, we tried out machine learning technique as an alternative and confirmed it works for several years data[2]. In this study, we apply the machine learning technique to 50 years of long-term muon data. We demonstrate that correction with machine learning technique works well by comparing to Mendoça’s method and by showing the variation amplitude is approximately proportional to the reciprocal of the median
rigidity.