Modeling Regional Interdependence in Business Data Based on Machine Learning Method

Abstract

In the business scenario, the level of product pricing or customer preferences is not only affected by the individual attributes of the product or customer, but also by the interdependency of other products or customers. For example, a customer’s preference may be more similar to its closer neighbor. This research introduces a complete system of models to study the preference interdependency among individual products or customers, so that the matrix for describing interdependence can be introduced into this model. This paper used the general Spatial Autoregressive Regression (SAR) to study these preference interdependency, which cannot be modeled with other standard machine learning methods. Moreover, an improved iterative lasso regression method is introduced to perform variable selection in the presence of interdependence. These models were illustrated by studying factors affecting Washington D. C. housing prices, where the prices are proved to be highly related to geographic networks.