Tingkat Keoptimalan Algoritma Pelatihan pada Jaringan Syaraf Tiruan (Studi Kasus Prediksi Prestasi Belajar Mahasiswa) Optimization Level of Training Algorithms in Neural Network (Case Studies of Student Learning Achievement Predictions)
| Main Author: | Sigit Sugiyanto, Hindayati Mustafidah Dimara Kusuma Hakim |
|---|---|
| Format: | Article info application/pdf eJournal |
| Bahasa: | eng |
| Terbitan: |
University of Muhammadiyah Purwokerto
, 2013
|
| Online Access: |
http://juita.ump.ac.id/index.php/juita/article/view/50 http://juita.ump.ac.id/index.php/juita/article/view/50/41 |
Daftar Isi:
- Abstrak - Algoritma pelatihan dalam jaringan syaraf tiruan telah diterapkan dalam permasalahan prediksi prestasi belajar mahasiswa Program Studi Teknik Informatika Universitas Muhammadiyah Purwokerto didasarkan atas nilai-nilai dalam mata pelajaran yang diujikan dalam Ujian Nasional saat di SLTA. Algoritma pelatihan yang digunakan adalah traingd, namun algoritma ini belum pernah dibandingkan tingkat keoptimalannya dengan algoritma pelatihan yang lain khususnya dalam kasus ini. Oleh karena itu, dalam penelitian ini dilakukan pembandingan tingkat keoptimalan dari algoritma pelatihan traingd, traingdm, learngd, dan learngdm dalam contoh kasus prediksi prestasi belajar mahasiswa Program Studi Teknik Informatika Universitas Muhammadiyah Purwokerto.Proses penerapan algoritma pelatihan dilakukan dengan pemrograman menggunakan bahasa pemrograman MATLAB. Sedangkan data hasil penerapan algoritma dilakukan uji perbedaanerror yang dihasilkan (uji keoptimalan) menggunakan perangkat lunak SPSS 16.0.Berdasarkan hasil uji statistik dari keempat algoritma diperoleh rata-rata masing-masing algoritma yaitu learngd: 0.0215, learngdm: 0.0163, traingd: 0.0211, dan traingdm: 0.0267. Dengan taraf alpha 5%, diperoleh nilai signifikansi sebesar 0,632. Hasil uji disimpulkanbahwa keempat algoritma pelatihan dengan beberapa parameter jaringan yang meliputi variasi epoh maksimum danlearning rate, tidak memberikan pengaruh terhadap tingkat error yang dihasilkan secara signifikan. Hal ini berarti tidak ada perbedaan tingkat optimalisasi atau keempat algoritma pelatihan memiliki tingkat optimalisasi sama. Kata-kata kunci- algoritma pelatihan, error, epoh, learning rate. Abstract - The training algorithm in neural networks have been applied to the problem of student learning achievement predictions of Informatic Engineering Study Program of Muhammadiyah University of Purwokerto that is based on the values in the subjects to be tested in a national exam on high school. The training algorithm used is traingd, but this algorithm has never compared the level of its optimization with other training algorithms, especially in this case. Therefore, in this study was done comparing the level of training algorithm optimization those are traingd, traingdm, learngd, and learngdm, in example of student learning achievement predictions at Informatic Engineering of Muhammadiyah University of Purwokerto. The process of application of the training algorithm was done using MATLAB language programming. The results of data application of the algorithms was tested using SPSS 16.0 to get error difference. Based on the results of the statistical tests of the four algorithms obtained an average of each algorithm are learngd: 0.0215, learngdm: 0.0163, traingd: 0.0211, and traingdm: 0.0267. By alpha level 5%, obtained the value significance of 0.809. This test results was concluded that the fourth training algorithms with some of parameter of the network that includes the variations maximum epochs and learning rates, not give effect on the resulting error rate significantly. This means there is no difference in the level of optimization of the algorithms or fourth training algorithms has the same optimization level. Key words- training algorithm, error, epoch, learning rate.