Predicting Student Dropout with Minimal Information

Abstract

Student dropout is a problem for both students and universities. However, in the crises that Lebanon is going through, it is becoming a serious financial problem for Lebanese private universities. To try to minimize it, it must be predicted in order to implement the appropriate actions. In this paper, a method to build the appropriate prediction system is presented. First, it generates a data source of predictor variables from student dataset collected from a faculty of economic sciences in Beirut between 2010 and 2020. Then, it will build a prediction model using data classification techniques based on identified predictor variables and validate it. Using open-source software and free cloud environments, a prediction program was developed. It consolidates, corrects, and normalizes the student's data. Then, it applies simple linear regression to show the correlation between the different variables and the student dropout, which allows us to select the factors that are highly correlated. From this point on, the program tries to predict the student dropout using different classification algorithms by machine learning on student dataset who left their courses either in success or in failure. Lastly, it measures the accuracy of the results and determines the best algorithm. In this study, the Artificial Neural Networks - Multilayer Perceptron showed an accuracy of 98.1% using only five variables. Finally, we evoke new avenues to further research and improve the model.