PREDICTIVE SALES MODEL USING MULTI-LAYER NEURAL NETWORK WITH BACKPROPAGATION ALGORITHM

Abstract

Accuracy is needed in prediction for efficient way to achieve profitability for trading firm and reduce costs. In this research, Multilayer Perceptron (MLP) of Artificial Neural Network with backpropagation is adopted to formulate a Multilayer Neural Network (MLNN) predictive model for sales of beer brand. Levenbreg marquardt backpropagation algorithm was used for training the network. Parameters chosen as input to the network were considered from weather, sales, seasonal and economic domain. After training and testing, the model was implemented with matlab and then compared with multiple regression analysis (MRA) and evaluated through the real world sales data provided by Nigeria Breweries PLC retail store. Experiment results indicate that MLNN preforms better than multiple regression analysis (MRA) in mean square error (MSE) and mean absolute percentage error (MAPE) measures.