The Implementation of Unsupervised Learning Techniques as a Data Sharing Model in the Back-propagation for the Classification of Student Graduation

Abstract

One of the requirements to increase the accreditation of higher education is the percentage of students who graduate on time. Responding to this issue, it is necessary to discover the factors that affect students in completing the Final Project. One of the algorithms that can be used to determine the classification of student graduation is the Backpropagation Neural Network (BPNN). The factors that have the most effect on student graduation, it includes procrastination, total credits and the number of repeat courses. To gain the best accuracy results on the classification technique, it was carried out by experiment of training and testing data sharing by applying the clustering technique. The cluster division consisted of the K-Means and K-Medoid algorithms, had the best cluster validity Davies-bouldin Index (DBI) 0.063 on the K-Means algorithm using 101 training data and 44 testing data. Based on BPPN, data sharing using K-Means had a big impact on the BPNN classification process with an accuracy value of 98% from a learning rate of 0.005 with 1000 iterations.