PREDICTING STUDENT’S FAILURE IN EDUCATION BASED ON DROPOUT STATUS

Abstract

Dropout is a significant issue facing universities in Indonesia. The decision to drop out is complex,
requiring consideration of various academic parameters or criteria. To address these challenges, leveraging data
mining techniques or machine learning in education can be an effective solution. Using the classification approach
with Neural Network (NN) methods to predict a student's academic status early can yield optimal results. Prior to
developing the NN model, supporting data undergoes pre-processing using methods such as the mean/average
method, z-score normalization, and information gain to determine the best parameters. Additionally, the Adam
optimizer is employed to fine-tune a parameter by iteratively updating weights based on training data. The
prediction model's performance is assessed using cross-validation as a benchmark. This approach achieves a
precision of 0.937. The most influential factors affecting dropout likelihood are grades, followed by failed courses,
student absences, and even the student's age.