ALGORITMA RESILIENT UNTUK MEMPERCEPAT PEMBELAJARAN BACKPROPAGATION PADA PERAMALAN DATA TIME SERIES

• Main Authors: , Dra. Wellie Sulistijanti, , Prof. Drs. H. Subanar, Ph.D
• Format: Theses and Dissertations NonPeerReviewed
• Published: [Yogyakarta] : Universitas Gadjah Mada 2012
• Subjects: ETD
• Online Access: https://repository.ugm.ac.id/100985/; http://etd.ugm.ac.id/index.php?mod=penelitian_detail&sub=PenelitianDetail&act=view&typ=html&buku_id=58000
• Institution: Universitas Gadjah Mada

Neural Networks training algorithm using the Resilient algorithm was developed to overcome the weakness of Gradient Descent algorithm to change the network weights and biases according to the bahavior of the gradient of each iteration of training by using only sign alone derivatives. Sign of this derivative will determined the direction of improvement of the weight, so the number of iterations required to achive the desired target less. This research uses data Seamarang city revenues between 2001 and 2009, best architecture Resilient algorithm for training MSE 0.01 is 1 input neuron, one hidden layer with 19 neurons and 1 output neuron (1-19-1) with parameters η + = 1,2 dan η � = 0,9. The use of Resilient algorithm for forecasting the best result in forecasting and has the speed, number of epoch is short and with simplified network compared to the Gradient Descent algorithm.