Data Mining In Applied Research, Sales Forecasting

The Enterprise Resource Planning (ERP) system is applied more widely when the level of enterprise informalization becomes higher and higher. The data stored in ERP is more tremendous day by day, and the data rich but information poor situation is increasingly obvious. Aiming at the disadvantages of sales forecasting in traditional ERP system and based on the characteristics of data mining technology, the author applied data mining to the system of sales forecasting in ERP.There are many factors such as basic demand, seasonal factors, periodic factors and stochastic factors, which influence sales. Traditional prediction methods only consider some factors, and the forecasting model is too simple to describe the complicated relations between factors. Artificial neural network (ANN) has good capability of nonlinear mapping and self-learning and it can solidify all the factors and the relations between factors into the sales forecasting model through learning. Therefore, based on back propagation (BP) algorithm, the model of sales forecasting is built. The prediction results indicate that the sales forecast precision is improved, the capability of generalization is good and the model has practical value.Although the model based on BP algorithm has good results, there are some shortcomings in BP algorithm such as local extremum points, forgetting the learned samples and slow rate of convergence. Then, genetic algorithm (GA) & Simulated annealing (SA) are applied to optimize the BP neural network. The optimization algorithm improves the rate of convergence and gets global extremum points. So the prediction results based on optimization algorithm are more effective than those based on single BP algorithm.