Research On Model Of Molybdenum Price Prediction Based On Decision Tree And Neural Network

Molybdenum is an extremely important non-ferrous metal resources,which is widely used and hard to be replaced,however its market price is affected by many factors that introduces strong volatility and sensitivity to the price,and also brings a lot difficulty to the prediction.Nevertheless,a prospective forecast of the trend of the price changes can make the enterprise take more effective and active measures to handle the situation,as well as making reasonable production and sales plan.Thus the value of resources will be highlighted,also the interests of the enterprises and nation can be better protected.The current methods for the analysis and prediction of the molybdenum price are generally unable to accurately describe its pattern of variation.Most of the scholars are only using mathematical modeling to find the law of price based on data within the past several years,or otherwise are only doing qualitative analysis of the trend of the future price of molybdenum by utilizing the factors that have influences on it.The results obtained from the former one normally lack of objectivity,and are not capable of dealing with mutative or unexpected situations,while the latter one are unable to reach a specific price range.Therefore,this paper,in view of the advantages as well as the disadvantages of the methods mentioned above,will explore the possibility of combining the two techniques to maximize the advantage:(1)The factors that restrict the price of molybdenum will be studied and classified according to their characteristics,and the relationship between these factors and the price variation will be analyzed.(2)The feasibility of using decision tree and neural network methods will be explored and analyzed to predict the price of metal,and an optimal algorithm will be introduced on the basis of two methods.(3)Upon the enormous factors that affect the molybdenum price,the representative and quantifiable factors will be selected,the sample set for the molybdenum price prediction will be built,and the price forecasting model will be established respectively according to the CART algorithm in decision trees,exhaustive CHAID algorithm as well as MLP,RBF and GRNN neural network algorithm.(4)The control variables method based on the same algorithm will be utilized to build models separately according to the different conditions of the model generation.(5)Using the model above to predict the price of molybdenum,the results from the same model will be firstly compared and analyzed.By drawing the comparative analysis diagrams of the prediction results and the actual price,as well as the comparative analysis chart of the error of the predicting results based on different models,with the statistical analysis of each error value,the best algorithm and model will be obtained for each class,and the final forecasting result will be given after the comparative analysis of the optimal results in the two respective methods.Empirical analysis shows that the decision tree and neural network algorithm can make more objective prediction of the price of molybdenum.Among those methods,the CART algorithm in decision tree and MLP neural network are able to predict the price with relative error rate no more than 2%.The method can not only realize the analysis of the importance of the factors that affect the price of molybdenum,but also draw more reliable predictive results.Meantime this mythology widens the application of the algorithm,and provides certain referential value for the research of forecasting the molybdenum metal price.