Artificial Neural Network Pada Industri Non Migas Sebagai Langkah Menuju Revolusi Industri 4.0

Abstract

The research conducted aims to make predictions with artificial neural metwork (backpopagation) and sensitivity analysis in the non-oil processing industry for the value of industrial exports. Data was obtained from the Badan Pusat Statistik (BPS) in collaboration with the Ministry of Industry of the Republic of Indonesia in the last 7 years (2011-2017). The process is carried out by dividing the data into 2 parts (training and testing) to obtain the best architectural model. The data processing uses the help of Matlab 6.0 software. Model selection is done by try and try to get the best architectural model. In this study using 7 architectural models (15-2-1; 15-5-1; 15-10-1; 15-15-1; 15-2-5-1; 15-5-10-1 and 15- 10-5-1) who have been trained and tested. By using the help of Matlab 6.0 software, the best architectural model is obtained 15-2-1 with an accuracy rate of 93%, epoch training = 189,881, MSE testing = 0.001167108 and MSE training = 0,000999622. The best architecture will be continued to predict the non-oil industry based on the most dominant export value using sensitivity analysis. From the architectural model a prediction of 5 out of 15 non-oil and gas industries contributes: Food & Beverage Industry, Textile & Apparel Industry, Basic Metal Industry, Rubber Industry, Rubber and Plastic Goods and Metal Goods Industry, Not Machines and Equipment , Computers, Electronics and Optics.