Comparative Study of Moving Average on Rainfall Time Series Data for Rainfall Forecasting Based on Evolving Neural Network Classifier

Abstract

Preprocessing is an important thing in data processing to get an important knowledge or a processing result with a good performance. Almost all classification or prediction process needs a precise preprocessing for its input data. In this paper, the preprocessing we're discussing is rainfall time series data smoothing using Moving Average algorithm. We used MA to observe the input vector for the prediction process using Evolving Neural Network (ENN). There are 4 types of Moving Average (MA) algorithm to be analyzed, which are Simple MA, Centered MA, Double MA, and Weighted MA. Modification with giving influence value to weight gained from weight function was done on Modified Weighted MA. After preprocessing were done, then those rainfall data will be going through forecasting process using ENN Classifier to get a 1-month rainfall forecast for Bandung Regency area, Indonesia. From experiments, the lowest MAPE was reached 15.66% from Centered MA or 84.34% accuracy for 1-month rainfall forecast.