Resumen
To examine how cross-strata neighboring behavior in a mixed-income community can influence the consumption choices of individuals in low-income groups, and to improve the prediction accuracy of the consumption choice model of low-income groups for small sample sizes, we developed a support vector machine (SVM) algorithm based on the influence of neighboring behavior. We substituted the predicted latent variables into the SVM classifier and constructed an SVM prediction model with latent variables based on reference group theory. We established the model parameters using cross-validation and used low-income residents from a mixed-income community in Shanghai as study objects to empirically test the model’s performance. The results show that the SVM selection model with latent variables has good prediction accuracy. The proposed model’s accuracy was improved by 1.29% on the basis of the particle swarm optimization (PSO)-SVM model without latent variables, and by 19.35% on the basis of the SVM model with latent variables. The proposed model can be employed to predict the consumption choices of individuals in low-income groups. This paper offers a theoretical reference for investigating neighboring behavior in a mixed-income community and the consumption choices of individuals in low-income groups and is practically important for urban community planning systems.