Backpropagation Neural Network for Predict Sugarcane Stock Availability

Abstract

National sugar consumption from 2013 to 2016 experienced an upward trend of 6.14%, especially for 2016, demand increased by 5.08% from the previous year [1]. Increment consumption of this sugar is inversely proportional to production results domestic sugarcane. Domestic sugarcane yields each year has decreased due to the problem of sugarcane on-farm or off the farm. Hence the government is trying to import sugarcane raw materials to fulfill national sugar needs. In importing raw materials sugarcane, the government needs predictive data for sugarcane production and the national sugar needs that will come used as a reference for the amount of raw sugarcane that needs to be imported. This study aims to estimate the results of sugarcane production based on historical data from 2017 to 2019. After making further predictions, it is necessary to evaluate the prediction model used. Method research used to predict production yield sugarcane, namely the Backpropagation Neural Network.Furthermore, the evaluation of the model used uses the Mean Absolute Percentage Error (MAPE). MAPE delivers a hint of how large the forecast error is compared with the series's real value. The following result 97% predictable production and 3% MAPE yield, then it can be said that the result of the method Backpropagation Neural Network has modeling capabilities decent forecast.

Penerapan Pemilihan Model Arsitektur Terbaik pada Neural Network pada Prediksi Jumlah Siswa SD di Kecamatan Siantar Barat

“…Arsitektur backpropagation merupakan salah satu arsitektur jaringan saraf tiruan yang dapat digunakan untuk mempelajari dan menganalisis pola dataset masa lalu lebih tepat sehingga diperoleh keluaran yang lebih akurat (dengan kesalahan atau error minimum) [2]. Terkait dengan hal tersebut, Metode Jaringan Saraf Tiruan Backpropagation banyak digunakan untuk menyelesaik an permasalahanpermasalahan terkait dengan prediksi [3]- [8]. Selain itu Backpropagation juga cepat, sederhana dan mudah diprogram [9], tidak memiliki parameter tuning selain dari jumlah input [10], fleksibel karena tidak memerlukan pengetahuan mengenai nework sebelumnya [11], [12] dan tidak perlu fitur khusus dari fungsi yang akan dipelajari [13]- [15].…”