Comparison Fletcher-Reeves and Polak-Ribiere ANN Algorithm for Forecasting Analysis

Abstract

Each method and algorithm ANN has different performances depending on the algorithm used and the parameters given. The purpose of this research is to obtain the best algorithm information from the two algorithms that will be compared based on the performance value or the smallest / lowest MSE value so that it can be used as a reference and information for solving forecasting problems. The ANN algorithms compared were Conjugate Gradient Fletcher-Reeves and Conjugate Gradient Polak-Ribiere. The conjugate gradient algorithm can solve unlimited optimization problems and is much more efficient than gradient descent-based algorithms because of its faster turnaround time and less iteration. The research data used for the forecasting analysis of the two algorithms are data on the number of rural poor people in Sumatra, Indonesia. 6-10-1, 6-15-1, and 6-20-1 architectural analysis. The results showed that the Polak-Ribiere Conjugate Gradient algorithm with the 6-10-1 architecture has the best performance results and the smallest / lowest MSE value compared to the Fletcher-Reeves algorithm and two other architectures. So it can be concluded that the 6-10-1 architectural architecture with the Conjugate Gradient Polak-Ribiere algorithm can be used to solve forecasting problems because the training time to achieve convergence is not too long, and the resulting performance is quite good.