Penerapan Backpropagation Neural Network Untuk Peramalan Penjualan Produk Susu
MetadataShow full item record
Milk product developed very rapidly nowadays. This is so many brand comes up to the market and makes competition between those products harder than before. Therefore each company must have analytical plan to predict how many product should be produced in order to avoid over product or lack of product and in order to keep distribution process run smoothly. Artificial neural network is one of machinal learning approach that can be used to solve so many problems primarily a complex problem than can not be modeled. Forecasting data time series is one of type of problem that can be solved by this approach. The objective of this research is to forecast milk product sale in KPBS Pangalengan, Bandung. This research consist of five steps such as problem identification, data collection and field information, preprocessing, neural network model, and postprocessing. This research use implementation of artificial neural network forecasting to predict milk product variant cup flavor in KPBS, Pangalengan Bandung. Artificial neural network consist of some composition data of milk product sale during January 2008-Juni 2010 that already trained based on work principle of human brain. Model of artificial neural network that used is this research is backpropagation neural network with two hidden layer and one output. Artificial neural network with one layer has a restriction in pattern recognition. This restriction can be solved by adding one or some hidden layer between input and output layer. Backpropagation is a model of artificial neural network with plural layer. Like other artificial neural network, backpropagation trains the network to balance its capability to recognize pattern and to give appropriate response to input pattern that similar to the pattern that used in training session. After training and examination, validation was done to evaluate the performance of artificial neural network in recognizing actual data pattern and the result shows that 4 data (67%) is appropriate to the target (actual data) and 2 data (33%) is not appropriate. This result shows that error in examination process is low. Mean square error (MSE) of validation process is 0.0286. After further examination bigger data training, MSE changed to 0.083. This MSE is bigger than the previous result. In addition, artificial neural network in this further research shows only 52% data that appropriate with target and 48% is not appropriate with real data. This result is affected by some factors. One of them is data that used in training and examination phase is random data between maximum level and minimum level of each attribute. The result shows that implementation of artificial neural network can help industrial sector to anticipate raw material and product stock accurately.