A course-level early final study status prediction task is to predict as soon as possible the final success of each student after studying a course. It is significant because each successful course accomplishment is required for a degree. Further, early predictions provide enough time to make necessary changes for ultimate success. This article aims at an effective solution to this task. Different from the existing works, we resolve the task in a more practical context. First, the temporal aspects of the task and its data are considered. For the task, historical datasets are used to support the task on current ones. For the data, both labeled and unlabeled data before the midterm break are used. Second, our solution examines assessment data for the task, and thus, requires less data collection cost and effort over time. Third, we propose a semisupervised learning method, ST_OS, to obtain a better prediction model because ST_OS handles data insufficiency when exploiting all the labeled and unlabeled data. Moreover, ST_OS combines Self-Training and Tri-Training to create a resulting ensemble model in an effective semisupervised learning process with local learning for each selected unlabeled instance. Above all, the task is addressed in a general manner for different course types. As a result, our solution outperforms several existing supervised and semisupervised learning ones with higher Accuracy and F-measure. Therefore, it can be used as a forecasting tool before their courses end. More activities can be then improved to help the students complete the courses successfully.