Improving the Efficiency of Forecasting Productivity, Using a Taguchi Experiment Design Approach (Case Study: Food Industries in Iran)

Productivity forecasting is a key factor in strategy planning in an organization. Artificial neural networks method is one of the productivity estimating methods whose users must have enough experience and skill because of its adjustable parameters. Trial and Error is mostly used to find the proper levels of these parameters. This article presents a seven step pattern for selecting proper adjustable parameters for neural network, using Taguchi experiment design method to improve the efficiency of productivity forecasting. As a result, the optimum parameters levels that lead to the most desirable forecasting in neural network are as follows: the number of hidden layers: 2 layers, the number of neurons in each hidden layer: 7 neurons, learning rate: 0.9 and the number of neural network inputs: productivity indicators with more than 0.85 degree of correlation. Among the above mentioned factors, the number of hidden layers with 71.18% of contribution rate in experiment results is the most important factor in neural network design to forecast the productivity of Iranian food industry. Finally, the overall results of the study showed that using this pattern provides the possibility of choosing competitive strategies besides decreasing forecasting time and cost. Moreover, this pattern helps decision makers with the extent of the consideration that must be put into each adjustable parameter by determining the contribution rate of each parameter in the experiment results.