When developing a black box model, the precise functional relationship between inputs and output is unknown. Engineers and scientists have turned to various regression tools in order to effectively capture the relationship based on past data observations. When modeling this data, however, it is important to only use inputs that provide information about the output. This paper presents a method of selecting the most informational input vectors for use in regression model building. This information-theoretic analysis for input vector selection requires only past data observations. Experimental results show that models built on the most informational input vectors produce less mean squared error on both training and validation data sets.

This content is only available via PDF.
You do not currently have access to this content.