Accuracy Analysis of ANN Back Propagation, Neuro-Fuzzy, and Radial Basis Function: A Case of HDI Forecasting
Syaharuddin1, Dewi Pramita2, Toto Nusantara3, Subanji4
1Syaharuddin, Department of Mathematics, Muhammadiyah University of Mataram, Mataram, Indonesia.
2Dewi Pramita, Department of Mathematics, Muhammadiyah University of Mataram, Mataram, Indonesia.
3Toto Nusantara, Department of Mathematics, State University of Malang, Malang, Indonesia.
4Subanji, Department of Mathematics, State University of Malang, Malang, Indonesia.
Manuscript received on October 01, 2019. | Revised Manuscript received on October 15, 2019. | Manuscript published on October 30, 2019. | PP: 1299-1304 | Volume-9 Issue-1, October 2019. | Retrieval Number: A9640109119/2019©BEIESP | DOI: 10.35940/ijeat.A9640.109119
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: One measure of the progress of a region or country is the increase in the Human Development Index (HDI) which includes life expectancy, per capita income, and old school expectations. HDI becomes an essential reference in time-series data, so it needs to be done forecasting process with reliable method. We use HDI data as much as the last nine years in West Nusa Tenggara province, which is one of the regions with the highest HDI acceleration in Indonesia in recent years. We do forecasting by comparing three methods namely Back Propagation (BP), Neuro-Fuzzy (NF), and Radial base Function (RBF), covering forecasting with 3 models of training and testing on the Back Propagation method, 9 training and testing models on A Neuro-Fuzzy method, and 1 training and testing model in the Radial base Function method. While the parameter accuracy (error) used in this forecasting is Mean Square Error (MSE). Based on the results of the simulation obtained NTB province predictions in 2019 using the Back Propagation (BP) method of 67.46 (increased by 0.23%); The RBF method amounted to 67.3 (fixed); and the NF method of 67.18 (decreased by 0.17%). From these results, the conclusion that in this case, the BP method is very good at doing simulation and decision-making results. The results were obtained from simulated data witt type training TRAINGDA, TRAINGDX, and TRAINRP. But simulation using type TRAINRP has the best parameter output with a performance (R) of 0.99194, a validation check of 1000, a gradient of 13.8, and a level of accuracy of 99.39%.
Keywords: Forecasting, HDI, Back Propagation, Neuro-Fuzzy, Radial Basis Function.