Research On Online Consumers' Purchasing Intention Based On LightGBM Algorithm

With the popularization of the Internet and the constantly development of mobile applications and electronic payment technologies,e-commerce has broken through the limitation of time and space.When consumers can consume anytime and anywhere,they also expect better shopping experience and personalized services.Common e-commerce marketing methods such as issuing coupons and product recommendations are all based on the analysis of consumer online behavior data.By analyzing the historical behavior data of users and accurately predicting their shopping intentions,it can further to provide consumers with targeted high-quality services,prompting their consumption to increase to purchase conversion rates.Therefore,how to effectively use consumers' behavior data and analyze their shopping needs is a challenge facing by all e-commerce companies.Based on the above research background,this paper aims to propose an online consumer behavior analysis system.The system uses consumer behavior data to build a data mining model,integrate the impact on different influencing factors of consumers' purchasing decisions,and finally predict users' purchase intentions.According to different shopping intentions,we can provide personalized services to customers and improve purchase conversion rate.The main research work of this paper includes:(1)Consumer behavior data preprocessing and feature engineering.The principle and implementation of outlier processing,feature selection and unbalanced data oversampling methods are mainly studied.(2)Build model.Model and adjust four candidate algorithms of support vector machine,random forest,gradient decision tree and LightGBM,and comprehensively evaluate four aspects including prediction accuracy.(3)The further research on the feature selection,cost-sensitive processing of unbalanced data and division of shopping intent of LightGBM model.The experimental results show that the cost-sensitive LightGBM model optimized by hyperparameters has the best prediction effect on e-commerce consumer behavior,with the AUC value of 0.893.Comprehensive evaluation shows that the model has good prediction accuracy,generalization ability and real-time performance for consumers' shopping intention,which meets the requirements of online e-commerce prediction systems to some extent.In addition,it is found that the clickstream data conveys important information about consumers' purchase intention.Using the classifier output prediction probability to further divide the user's shopping intentions,it can initially locate loyal users,potential users and churn users,and then develop targeted marketing strategies to achieve precision marketing.Through the above research,this paper enriches the theoretical basis of purchase intention based on consumer behavior data,provides a better model for online consumer behavior analysis,and has certain theoretical and practical significance for the establishment of a realtime online consumer behavior analysis system.