基于Bagging与决策树算法的在线拍卖成交价格预测模型

通过分析在线拍卖出价特点,利用决策树和Bagging算法建立了一种全新的在线拍卖成交价格预测模型.作者编写程序收集淘宝网在线拍卖交易数据3310条,对应有效出价记录8275条. 数据分析表明, 如不考虑未成交商品,则有40.4{\%}的交易可以利用出价次数精确计算最终成交价格.如将未成交商品视为成交价格为0, 该比例可提高为79.55{\%}. 据此发现,作者通过预测出价次数间接对成交价格进行预测. 实验证明,模型明显优于平均值预测, 并有21.7{\%}的预测结果完全准确.通过与Heijst发表于《Decision Support Systems》上的研究进行对比,结果表明预测模型在样本需求量、运算时间,及完全准确预测率上有明显优势. 由于模型训练时间仅为数秒,为建立实时在线拍卖成交价格预测决策支持系统奠定了基础.

Online auction final price forecasting model based on Bagging and decision tree

By analysing bidders' behaviors, the author proposed a new model which is based on the Bagging arithmetic and decision tree for predicting final prices in online auctions. The author collected 3310 transaction data and corresponding 8275 bids from Taobao. Data analysis shows that the final prices of 40.4% transactions can be calculated by using the times of bids. If the transactions which have no bid are included, the percentage can increase to 79.55%. Instead of predicting the final price directly, the author chose to predict times of bids first and then used it to calculate the final price. The experiment proves that the model substantially outperforms the naive method of predicting the category mean price, and 21.7% of predicted results are exactly equal to the real ones. The author also compared the model with Heijst's research which was published in Decision Support Systems. The result shows that the model is better in required training sample size, calculating time and the percentage of accurate prediction. For the training time is only a few seconds, our research can lay the foundation for developping real-time dectsion support systems.