Rainfall prediction in Kemayoran Jakarta using Hybrid Genetic Algorithm (GA) and Partially Connected Feedforward Neural Network (PCFNN)

The weather changes easily these days, that is difficult to predict. Yet, weather forecasting is the important and useful thing in all aspects of life for instance, in the agriculture field to decide the time of planting. Thus, weather forecast of rain fall intensity particularly in region of Kemayoran Jakarta is conducted in this research. The forecast system built uses Hybrid Genetic Algorithm (GA) and Partially Connected Feedforward Neural Network (PCFNN). In this research, optimum weight and connection neural is optimized with the type of PCFNN. Previously, it has been conducted a research using Evolving Neural Network with all connected neural network and the result was MAPE prediction result rainfall 38.82% or accuracy as 61.18%. Yet, based on the experiment hybrid algorithm between GA and PCFNN, it is gained the MAPE value 35.20% or accuracy as 64.80% with the solution of data missing value in rainfall intensity replaced by 0 (zero) and MAPE 18.48% or accuracy 81.52% for missing value solution in rainfall intensity replaced by mean value.

I. Introduction

Previous years, dry and wet seasons in Indonesia, especially in Kemayoran Jakarta could be predicted by only using the calendar, but since the weather system is getting uncertain and can cause disadvantages of many fields, daily activities and many others, thus it is needed a system which can accurately predict the weather changes. In the previous research, it had been building a weather forecast system which used Artificial Neural Network (ANN) with the optimum of Conjugate Gradient, but the result from the research couldn't be used predict the rainfall because of the high Mean Absolute Percentage Error (MAPE) [1], [3]. On the other hands, the research on rainfall prediction using algorithm Evolving Neural Network (ENN) which has fully connected feedforward neural network [3], can predict the rainfall with MAPE as 38.82%. in this research, it is hoped that with the implementation of hybrid Genetic Algorithm (GA) and PCFNN which has partially connected feedforward neural network, enables us to produce a better rainfall prediction.