Data Mining and Artificial Intelligence: Februari 2017

Jumat, 10 Februari 2017

Predicion and Anomaly Detecton of Rainfall for Planting Time

Based On Evolving Neural Network in Soreang, Bandung

Gunawansyah

Supervisor : Prof.Dr. The Houw Liong

Co-Supervisor : Prof. Dr. Adiwijaya
/>

As an agricultural country and located around the equator line, Indonesia

geographical position between two continents and oceans is very sensitive to

regional and global atmospheric circulations as sunspot, cosmic rays, Indian

Ocean Dipole(IOD) and Southern Oscillation Index(SOI). The disruption of one

circulation can affect climate and weather. Extreme anomaly climate is very

influential in the agriculture field because it can decrease growing area,

production and productivity of food crops. In this research, rainfall

prediction has been conducted using Evolving Neural Network (ENN) and

completed with anomaly detection for finding the best starting time for

planting, so the risk of loss due to rainfall anomalies can be minimized

because farmers can adjust the planting starting time with the change of the

rainfall and the attention to anomalies that occur during the growing seasons.

From three scenarios, one hidden layer in Artificial Neural Network (ANN)

architecture was sufficient and ENN had good performance in different dataset.

Rainfall prediction result used all data (January-December) from 1999-2013 had

the accuracy of 84.6%, 66.02% for dry season (April-September) and 79.7% for

wet season (October-March). Based on prediction and anomaly detection in this

research, in Soreang the first week of January, April and October 2014 we

recommended for the starting time of planting in 2014.

Keywords: Rainfall, Prediction, Evolving Neural Network, Genetic Algorithm

Artificial Neural Network, Anomaly Detection

Data Mining and Artificial Intelligence