Detection of Potentially Students Drop Out of College in Case of Missing Value Using C4.5

The reputation of a university can be determined by the number of students drop out. This problem was experienced by many universities in Indonesia. It has been conducted by many researchers, however the data acquisition, attributes were not well explained. This study is aiming for giving projection related to the reasons behind students drop out by using machine learning technique. The challenging phase of preprocessing primary datasets are missing value, balanced class distribution, and a variety of data types. Two classes are applied: drop out and graduate students. By analyzing the problem of missing value data, it can reflect the basis of why students drop out or students who have the potential to drop out. According to the problem of balanced class distribution, Decision Tree algorithm is utilized, meanwhile for tackling the various of data types, we use C4.5. The result shows that 20 attributes using stratified sampling is the best of among all datasets and experimentations with an average AUC, accuracy, precision, and recall values of 0.98, 96.87, 98.75, and 97.84 respectively. It indicates that the proposed method is suitable for predicting students drop out with a balanced case of class distribution, despite having a missing data value problem.