A comparative study of linear regression and decision trees for predicting consumer purchase intention across consumption scenarios of different complexity

With the development of e-commerce, social media marketing and digital operations, enterprises can obtain a large amount of consumer data such as prices, revenues, discounts, product ratings, browsing frequencies and historical purchase records. How to utilize these data to predict consumers' purchase intentions has become an important issue in business analysis and precise marketing. This paper compares the prediction characteristics of linear regression and decision trees in different complexity consumption scenarios. Through literature research, simulation data analysis and comparative analysis, two types of experiments are constructed: the first type only sets an approximately linear relationship between product prices and purchase intentions; the second type adds income, discounts, ratings, browsing frequencies and historical purchase frequencies, and sets threshold effects and variable interactions. The research results show that in simple linear scenarios, linear regression can achieve slightly better prediction performance with lower complexity; in multi-variable, non-linear and interactive relationship scenarios, the prediction error of decision trees is lower. The paper believes that the selection of business models should be based on data structure, prediction accuracy, interpretability and implementation cost, rather than simply pursuing model complexity.