Prediksi Prestasi Mahasiswa Berdasarkan Nilai Ujian Nasional Menggunakan Jaringan Saraf Tiruan dengan Metode TrainBP dan TrainBPX

Abstract

Prediction of student achievement in this research only predict based on national test scores by using the trainbp method and the trainbpx method. This research uses the input patterns of the subjects tested on the national exam for high school students, namely: Mathematics, Indonesian language, English, Physics/Economics, Chemistry/Sociology, and Biology/Geography, while the pattern of the desired output or target is divided into two patterns, namely the pattern (0) assuming no student achievement and patterns (1) assuming achievement students. Then these values are trained and tested for the system to recognize it properly. After training and testing, the neurons that are considered worth more than 0.5000 reputed well worth 1 (achievement), whereas neurons that are considered worth less than 0.5000 reputed well worth 0 (no achievement). The results of the training trainbp method and trainbpx method of 150 training data with the results reached 100%. Test results of the testing (outcome prediction) of 100 with trainbp method with the results of 33 data (33%) achieved a prediction accuracy (in accordance with the target) and 67 data (67%) did not correspond with the target, whereas with trainbpx method, the data showed 69 (69%) achieved a prediction accuracy, and as many as 31 data (31%) did not succeed, because incompatible with the target.