Date of Completion

Spring 5-15-2020

Thesis Advisor(s)

Ming-hui Chen

Honors Major



Applied Statistics


The Box-Cox transformation is a way to transform non-normal data into more normally distributed data. However, when we fit linear regression models to transformed data, we cannot use the Akaike Information Criterion (AIC) directly to compare different models since the transformed data are no longer on the same scale. In this study, the Jacobian adjusted AIC is proposed to compare regression models on transformed data and to select an “optimal” value of the transformation parameter. Instead of using a single for the whole data, which is commonly used in the literature and in practice, we formulate a linear regression pattern so that the transformation can be adaptive to the changes in the data. The proposed method will be applied to fit and analyze different types of energy data including steam, chilled water, and electricity on the University of Connecticut Storrs campus.