B2C Potential Customer Identification Research Based on Random Forest
This paper proposed a potential customer discovery method from the perspective of customer segmentation, using factors of consumer behavior and combining Random Forest with B2C website web log files to identify the potential customer from B2C website visitors.Firstly, we got data of users behavior according to a series of processes of data acquisition, data processing and analysis.Secondly, we analyzed the effects of the variables number and trees number to the consequence of random forest model by setting several groups of repetitive experiments, thus we could confirm the most optimal model parameters the best model.Then, we verified the accuracy of model by means of ten-fold cross-validation experiment and recall rate experiment, etc.Compared with other models,random forest model shows the best.Finally, we made a further validation to the identification results with real data.The models experimental results show that the number of viewed pages and average page browsing time are the most important factors of the consequence of the model.At the meantime, the model identification obtained an optimal prediction accuracy.The results of this study has a significant guidance to the development of B2C e-commerces marketing strategy.
data mining customer segmentation potential customer random forest
WANG Danshi
School of Economics and Management, Nanjing University of Science and Technology, Nanjing 210094, China
国际会议
武汉
英文
250-255
2016-10-21（万方平台首次上网日期，不代表论文的发表时间）