Modern Regression Methods: A Comparative Discussion

In the last ten years, several new piece-wise linear analytical methods have been developed and applied to problems of prediction and function approximation with marked improvement over traditional methods. This progress is important for applications in any field which have in the past used standard linear procedures or their variations. These new procedures should be considered wherever one might be considering discrimination or classification procedures, multiple regression, time series analysis and forecasting, econometric model building, and neural modeling. Examples abound in Marketing, Finance, Accounting, Information Systems, Economics, Engineering, and Mathematics as well as many other sciences. There are differences between these procedures that are not widely understood, and their relationships to more commonly used methods have been discussed only in widely scattered articles in the literature. This paper discusses several of these new developments within a common framework, describes applications and potential applications in various areas, and discusses sources of computer software for interested users. The well-known Boston Housing data is treated with each of three procedures for comparitive purposes.