The Final Transaction Price Forecasting Based On The Online Auction

Because of the popularity of online auction,online auction commodities of the final sale price prediction is particularly important.Through the forecast,the tenderer and sellers can get useful information,dynamic and rules,better grasp the auction to run auction.For the purpose of this article,we use Ebay auction history data on final sale price is forecasted,and the final sale price is divided into continuous and discrete price prediction.For continuous price prediction,in order to eliminate the correlation between attributes,it is necessary to construct new attributes first,then select all attributes with Multiple stepwise regression,and finally select the attributes that have a greater impact on the final transaction price,and establish the prediction model with regression t ree.The final sale price is divided into contains only 300 dollars in the following sections and include all predict clinch a deal the price of two parts,respectively,we found that this can improve the model accuracy and no loss of useful information of the data set.The article adopts the root mean square error as the measure and the several popular machine learning algorithm,this article chooses multiple linear regression model,multi-layer perceptron model,K nearest neighbor type mold)are compared;For discrete price forecast,we will it as a classification problem,the use of extreme gradient ascension(XGBoost)set up model and forecasting,at the same time the model with Logistic Regression model,and Logistic Regression,comparison,used as a measure accuracy,through after 10 iterations,take the last time the accuracy of the final results for the model accuracy,finally it is concluded that the selection of model in the actual auction data has a higher accuracy.