The importance of retention rate for higher education institutions has encouraged data analysts to present various methods to predict at-risk students. Their objective is to provide timely information that may enable educators to channel the most effective remedial treatments towards precisely targeted students in an efficient manner. The present study, motivated by the same demand, proposes a deep learning model trained with 121 features of diverse categories extracted or engineered out of the records of 36,917 postsecondary students from 2006 to 2019. While this vast sample set makes the model more realistic, the diversity and abundance of features make it more likely to identify influential causal factors that educators can further manipulate. The model undertakes to identify students likely to graduate from a university of interest, the ones likely to transfer to a different school, and the ones likely to drop out and leave their higher education unfinished. The present study is the first of its kind that accepts the challenge of labeling transfer students. Since the cost of dropout and transfer grows by the time a student spends in a school, these predictions will be performed in different stages of education. The time aspect is accounted for by incorporating LSTM in the model. Our experiments demonstrate that distinguishing between to-be-graduate and at-risk students is reasonably achievable in the earliest stages, and then it rapidly improves; but the resolution within the latter category (dropout vs. transfer) depends on data accumulated over time. However, the model remarkably foresees the fate of students who stay within the school for three years. The model is also assigned to present the most impactful features with respect to each of the three categories, both on institutional and individual levels. Furthermore, we adjust the model for different predictable scenarios in which, for example, transfer students could be further classified based on the level of the program to which they transfer. In this respect, this dissertation presents a preliminary study that proposes a hybrid CNN-BiLSTM model that successfully identifies the level of programs in the student-tracking records collected nationwide, which suffer from inconsistencies in formatting. As a potential application, the predictions made by our model can support a forecast system that exposes each student to the personalized projection of her/his final performance. This dissertation presents a case study on two online core courses to illustrate the contributions of such forecasting through altering students' learning strategies and enhancing their engagement, whose impacts are independently investigated. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com.bibliotheek.ehb.be/en-US/products/dissertations/individuals.shtml.]