In order to understand the learning patterns of middle and high school students in the second semester of 2020, when in-school and remote classes were conducted in tandem due to the COVID-19 situation, this study identified learning types according to self-directed learning, explored the predictive variables that affected them, and compared differences in class attitudes, self-efficacy, and academic emotions between groups. To this end, 783 middle and high school students in Seoul were subjected to latent profile and polynomial logistic regression analyses. The results are as follows. First, three latent classes were derived: an "average group for self-directed learning," which accounted for more than half of the number of students in both middle and high school; a "high group for self-directed learning"; and a "low group for self-directed learning." Second, the better the self-directed learning group, the better the self-efficacy and class attitude of both middle and high school students. In addition, high school students in the average or lower group felt more anxiety and stress than before the COVID-19 restrictions than did those students who were good at self-directed learning. Finally, it was found that the variables that impact latent class classification for middle school students were out-of-school classes, self-efficacy, parental support, and the ability to use digital devices, while for high school students, they were self-efficacy and parental support. These results could provide the basis for the development and operation of student support programs focused on the teaching and learning process in relation to learning recovery made necessary by the COVID-19 pandemic.