Model Identifikasi Peta Secara Otomatis Menggunakan Konsep Jaringan Saraf Tiruan Backpropagation

Abstract

Artificial Neural Network (ANN) is a model from many models in Artificial Intelligence field. ANN mimics the behaviour of human brain concept of learning. In this model, information will flow and distribute from one neuron to another neuron by using huge of computational process. Backpropagation is the one famous model in ANN. Most researcher in Indonesia using this model for forecasting and deciding data in many field of application like economic, natural science, law, psychology, social etc. The concept of backpropagation is a supervised model, it means inputs and outputs should decide and hold first before trained. Every point of neurons connection from one layer to another layer has a value associating by its weight value.  This paper will introduce the behaviour of ANN model to recognize every place of region in map automatically. The case, we will take one sample region map from Daerah Istimewa Yogyakarta Province. After learning and training from 420 data, we have got a stable weight of all neuron connections, than from this ideal weight of connection we will continue testing every map segment by using more than 20 testing data. The result shows that accuracy at 80% of testings with less than 20% failure. The best ANN architecture is found with the following parameters: 0.02 learning factor, 0.01 threshold failure, 10 hidden units, and one layer of hidden layer.