Jaringan saraf tiruan untuk prediksi tingkat kelulusan mahasiswa diploma Program Studi Manajemen Informatika Universitas Negeri Gorontalo

Abstract

In this research is made a system of student graduation rate prediction using artificial neural network and back propagation method. Predicted graduation rate is length of study and grade point average. The Input variables is the value of 16 subjects in the first year program. Output variables are 2 categories for the length of study and 3 categories for grade point average. The aims of this research are to get the best parameters and architecture of artificial neural network, to predict student graduation rate and to measure the input variable sensitivity. To obtain a convergent learning outcomes, has done some trial and error with several variations the number of hidden node, learning rate and training function, to generate the high level of data. The result shows, graduation rate prediction using artificial naural network with back propagation method has good result, which 100.00% validation data generalization and 93.94% testing data generalization for lenght of study prediction model, and 96.67% validation data generalization and 100% testing data generalization for grade point average prediction model. The best parameters and architecture from the both of prediction models generated by training in second data group and levenbarg-marquardt training functions. Length of study prediction model, optimal on learning rate 0.1 and number of hidden node 10, and grade point average prediction models optimal on learning rate 0.05 and number of hidden node 15. The sensitivity analysis results that subject as input variables, which have the most impact is Aljabar Vector and Matriks.