Design Of Predictive Model Of Commodity Purchase Behavior Based On Machine Learning

In recent years,more and more e-commerce companies have built their own personalized recommendation systems on shopping platforms.The so-called personalized recommendation refers to the marketing mode of recommending different products to different consumers according to their own preferences.At present,machine learning has been widely used in the field of big data,and the design of personalized recommendation system just needs to rely on massive interactive data between consumers and commodities.therefore,it is feasible to apply machine learning to recommendation system.This article uses Ali's historical record of consumers and commodities in a certain period of time to judge consumers' "favorites" and provide important basis for the future construction of a personalized recommendation system.The main tasks completed are as follows:1.A data preprocessing method based on time series is designed and implemented.First,the "noise" data is filtered through data cleaning.Then,according to the characteristics of time series of the original data,a data "slicing" mechanism is designed and the data is " sliced" processed;2.Through analyzing the collected data and combining with the business logic,designing a derived feature group,extracting and transforming the data to obtain high-dimensional trainable sample data;3.Creatively combining the stability selection method with Pearson correlation coefficient method to obtain a new feature selection algorithm SSP,and finally using SSP algorithm to select features of samples;4.A "uniform downsampling method" based on time series is proposed,which balances positive and negative samples with cross-validation.Then,the XGboost model is adjusted with the grid search method before training;5.Aiming at the problem that a single model is too sensitive to data and easy to over-fit,a XGboost hybrid model based on bagging strategy is constructed by using bagging integrated learning idea.Finally,through two sets of comparative experiments,this paper proves that SSP algorithm can effectively screen features,simplify the complexity of the model to a certain extent,and improve the classification accuracy of machine learning.Compared with a single machine learning model,the XGboost hybrid model based on bagging strategy is not easy to lead to over-fitting and thus has stronger robustness.