The prediction of transport properties in Josephson junctions based on first principle and machine learning