ABSTRACT
This chapter introduces linear regression models to predict the fair market values. It provides many interaction terms to improve the prediction accuracy. With so many possible interaction terms, the chapter employs overlapped group-lasso to select the important interaction terms. The first-order interaction model has a simple form. However, learning interactions is quite challenging, especially when there are many explanatory variables. The chapter analyses the performance of linear models with interactions based on representative policies selected by Latin hypercube sampling. It discusses the use of linear models with interactions to predict fair market values. The results shows that linear models with interactions outperform models. The results indicate that it is important to model interactions when developing models for predicting fair market values. The validation measures in the output show that the results are very good but not as good as the case when conditional Latin hypercube sampling was used to select representative policies.
