Optimalisasi Parameter dengan Cross Validation dan Neural Back-propagation Pada Model Prediksi Pertumbuhan Industri Mikro dan Kecil

Abstract

It is important for us to predict what will happen in future and to reduce uncertainty. Various analyzes are therefore necessary in order to optimize or improve the prediction results by several methods. The objective of this research is to analyze predictive results by optimizing the training and testing by means of cross validating parameters on the growth of micro and small-scale production in Indonesia through the exactness of the return-propagative method. The method of reproduction is used. These results are compared with results of backpropagation during training and testing without optimisation of the same architectural model. The dataset is based on the growth in production in micro and small businesses by province from the Central Statistical Agency(BPS). There were 34 records in which data from 2015-2019 for growth of production were collected. The results with optimisation have surpassed without optimisation the back propagation model by looking at RMSE, in which the best RMSE in the 3-2-1 architectural model was obtained and the side type is mixed sampling. The obtained RMSE value is 0.1526, or a difference between the best background architectural model, 3-2-1 and 0.0034. (0.157). The results of this model were 94 percent.