The number of dropouts from universities is increasing, and many researchers have claimed that poor academic grade is a major reason that they drop out from the universities. Therefore, there is a need to develop a system that can forecast the post-enrollment grades of the students at an early stage and help the students with poor academic grades. This paper aims to propose a method that forecasts whether the rank of Grade Point Average (GPA) after a student enrolls in a university is lower in the department based on machine learning using the feature values, such as his/her study logs on the pre-admission educations of the university, his/her academic grades in high school, and the data on the admission examinations s/he took. While previous studies have conducted the experiments to forecast academic grade 1 year later, this paper conducts an experiment to forecast academic grade at the end of each semester of the university.