Automation of Multiple Linear Regression Analysis Using Neural Networks

Roshan D. Yapa, Ajantha S. Atukorale


Statistical regression analysis is a powerful tool that can be used to model relationships between variables in a statistical data set. Also, regression analysis is a widely used technique in the statistical data analysis. Regression analysis needs an expert knowledge on the subject Statistics. But most of the people, who need regression analysis to be done for their existing data, do not have enough background knowledge on Statistics. Due to this reason, they are required to get an assistantship from a statistician. Even though enough statistical software packages are available for regression analysis, the person who is operating that software is also expected to have knowledge on Statistics for selection of suitable variables, selection of suitable transformations, interpretation of results etc. This is a great disadvantage for most of those people. So far few attempts have been made to automate regression analysis using neural networks. But all of those attempts have been made to get improved regression estimates using neural networks rather than addressing the above problem. This paper focuses on how neural networks can be used to automate regression analysis and simplify the regression estimation process. Also this paper introduces a neural network model for multiple linear regression analysis.