EPM model research based on intelligence decision tree and BPNN
Weihua Niu · Jisuanji gongcheng yu sheji · 2008
Due to the data is tremendous, jumbled and bad quality in EPM field, so it is exigent to construct high accuracy forecast model. Intelligence decision tree classification algorithm is adopted, the datum is classified according to their attribute values. The attribute value is replaced by arithmetic average of all attributes which class of half-baked data belongs to and the non-attribute value is replaced by the highest appearance frequency of non-attribute in the same attribute. By this way, half-baked datum which will enter the EPM model is supplied and data optimization is ensured. A new method of using BPNN in EPM is introduced. BPNN computational complexity is reduced by improving hidden nodes of BPNN. The experimental result indicates that the model has good predicting effect.