Optimization of Weight Backpropagation with Particle Swarm Optimization for Student Dropout Prediction

Abstract

Students who drop out of school are the cases that should be of concern in college. An uncontrollable decline affects the quality of the university. Dropout could happen for a variety of reasons, one of which has carried out a maximum study period. The Undergraduate program has a maximum finished within less than eight years. In this study, student dropouts prediction was conducted for students who had the possibility of exceeding the maximum study period. The predictions are made by digging the data patterns in the student’s academic database by utilizing the Academic Achievement index data of each semester and class attendance. This research aims to accelerate training and improve the accuracy of predictions with backpropagation (BP) optimized with particle swam optimization (PSO). Evaluation of the classification model is done with 10 fold cross-validation, where the data are divided into 10 fold, and the test is done 10 x until the highest accuracy is obtained. Further accuracy results was compared with a simple backpropagation algorithm without optimization. The algorithm of backpropagation, which is optimized with particle swarm optimization, produces the best accuracy of 100% with 6 epoch. Meanwhile, the backpropagation algorithm generates an accuracy rate of 77.78% with 17 epoch. The highest accuracy gained using the best architecture is 8-8-1 and the number of particles 10. Optimization of a network weight backpropagation with particle swarm optimization can improve the accuracy and number of decreased iterations