Neural Network -Based Prediction Model for Determining Student Expertise

Abstract

This study uses a neural network algorithm approach with backpropagation method to predict the accuracy level of determining the student's area of expertise. Neural network algorithm is an artificial neural system or artificial neural network is a physical cellular system that can obtain, store and use knowledge gained from experience, to activate using sigmoid bipolar where the output value ring is between -1 to 1. This area of expertise or concentration will taken by students through the selection of subjects of expertise that are in the curriculum of the study program offered. Academic supervisor / academic guardian who will direct students in choosing subjects according to the study interests desired by students. However, this direction is not well known by students. As a result, there is a possibility that students do not choose their field of expertise properly and it will affect the final score of the subject of expertise and competency testing. The analysis technique in this study uses descriptive statistical analysis which is an analysis of the frequency distribution, the size of the concentration, and the size of the spread. Because this study uses a neural network algorithm approach, the frequency distribution, the size of the concentration and the size of the spread are the number of hidden layers, iterations, learning rate, MSE and the confusion matrix (accuracy), which is to know the level of accuracy. The data to be analyzed is maining data or data which is training data consisting of 12 criteria (inputs) and 1 output (consisting of 3 areas of expertise), this data will be integrated into the Alyuda Neurointelligence software , from this software the number of hidden layers is determined , iterations, learning rate , MSE. The results of this study obtained the level of prediction accuracy model, namely: 84.8484% (hidden layer 2, neuron input layer = 13, learning rate = 0.1 and iterations = 500), 91.9191% (hidden layer 2, neuron input layer = 13, learning rate = 0.1 and iterations = 1000), and 95.9596% (hidden layer 3, neuron input layer = 13, learning rate = 0.1 and iterations = 2000).