Household Consumption Prediction Based on Machine Learning Models: A Case Study of the U.S. Pet Economy

Abstract

In all developed and developing economies, the pet economy has emerged as an important component of household consumption in the recent period. Yet, many studies analyze just one dimension of spending by just one predictive approach. This paper fills that gap by investigating the performance of three predictive models, including linear regression, random forest, and XGBoost, across three levels of the U.S. pet economy. These levels include aggregate industry expenditure, household spending by household size, and veterinary costs by species and visit type. Every model type corresponds to one of the data structures which can in theory fit best. The change in aggregate expenditure, between 2018 and 2026, is explained by linear regression 95.3 percent. Random forest reflects a pattern which is not monotonic, with three-person household groups spending more than other household groups. XGBoost identifies a clear threshold-based difference between routine and surgical veterinary costs. Results indicate that a superior performance is not achieved with a specific model, but with a particular data structure. There are different patterns of aggregate, household and medical spending. The findings provide a structural framework for forecasting demand in this growing consumer sector.