Predictive Modeling of Graduation Outcomes in Islamic Boarding Schools Using Feedforward Neural Networks

Abstract

This study investigates the application of Feedforward Neural Networks (FFNNs) to predict the graduation status of prospective students at Pondok Pesantren Nuur Ar Radhiyyah, an Islamic boarding school emphasizing Quranic memorization, religious practices, and prayer as core educational values. Manual analysis of student test results for admission is becoming inadequate due to the increasing number of applicants. Previous research has successfully applied FFNNs to predict student graduation rates in various educational settings. This study explores the impact of three activation functions (Sigmoid, ReLU, and Tanh) and the number of hidden layers and neurons per layer on FFNN model performance. A dataset of 480 prospective student evaluations encompassing educational level, Quranic memorization score, religious practices score, prayer score, and graduation status was analyzed. Twelve FFNN models were configured with different activation functions, hidden layers, and neurons per layer. Model performance was evaluated using 10-fold cross-validation. The results revealed that the model utilizing the Tanh activation function with four hidden layers and four neurons per layer achieved the highest accuracy (97.6%), precision, and recall rates. This research highlights the potential of FFNNs for predicting student graduation outcomes in Islamic boarding schools and emphasizes the importance of activation function selection and hidden layer architecture optimization for achieving optimal performance. The complete dataset is available on Kaggle.com for further research.