Research On Sales Forecasting Model Based-On Rough Set And Neural Networks

Marketing is a highly complex nonlinear dynamic system, its change have their own regular pattern, and also influenced by political, economic and psychological factors. There are many difficulties when traditional quantitative forecast method based on mathematical statistics in the research of marketing.But the neural networks have some characteristics, such as self-organizing and adaptive, can extracted the knowledge related to economic activities from historical data. It is very suitable for solving some problems of sales forecasting. A large number of simulation results show that the neural network have a certain practicability in the sales forecasting application.Neural network utilizes the use of nonlinear mapping and parallel processing, and use the structure of itself to express the implicit functions coding of relationship knowledge between input and output. The mapping relationship between input and output space learn and adjust continuously by the structure of network, finally determine the specific structure expression of network.,realize supervised learning.But it can not handle input data of semantic form and simplified information space dimension.When the input information space dimension is a larger scale, it will lead to complicate the structure of network and make the training process need too much time.Rough set theory was based on some concepts, such as attributes dependence, reduction, nuclear, rules extraction and distinct matrix, it can retrench knowledge system, realize the pretreatment of information system,remove redundant attributes and redundant samples, compressed information space dimension,etc. A rough set and neural network forecast model was proposed by the analysis of these two methods.Rough set theory be used as pre-system,it can mine several more important factors from the commodity sales data. It also can obtain a retrench attribute set under the premise of without decreasing data consistency as much as possible. This can reduce the input data of neural network and without affecting the network capability about incident detection. So as to reduce the complexity of the neural network and its training time. Meanwhile improve the speed and accuracy of forecast. This paper put forward the model and learning algorithm based on network and rough set theory. The model has fast learning and fault-tolerant capability characteristics. So it improves the neural network's forecasting accuracy as well as reducing its burden of learning. In order to obtain better accuracy of the forecast, the paper also uses the combined neural network.Appling this model to forecaste sale quantities of a liquor enterprise, and carry on comparative study and demonstration analysis with each single forecast model.Finally, it summaries the result to the thesis, and put forward the further research work.