1308100110-abstract en

NEURAL NETWORK BACKPROPAGATION ALGORITHM ON MODELING EAST JAVA POVERTY INDICATORS Name NRP Departement Advisor

: Eka Prasetiyo : 1308 100 110 : Statistika FMIPA ITS : Dr. Bambang Widjanarko Otok, M.Si ABSTRACT

Modeling input variable against output one is either matter that often be done on statistic analysis. This research is purposed to build a model using neural network method. It uses poverty data SUSENAS 2010 and poverty information data regency or city 2010. The data consist of ten inputs and an output that is poverty depth index. The Analysis is to compare the effect of combination of the number of hidden layer, learning rate value, and replication on neural network backpropagation algorithm. The result shows there is no effect when using a node hidden, but it will give an effect if using two or three hidden. That cause the value of RMSE and the percentage of significance parameters is distinct. Next analysis chose the best model every replication, learning rate and hidden layer node. The best model is three nodes of hidden layer and learning rate equal 0.005 or it is completely as MLP [10,3,1]. The influence variables on model are toddler’s birth assisted by medical personnel, latrine facility, pure water facility, and the literacy rate. Keywords: Hidden layer, poverty, learning rate, neural network, RMSE

v