Neural networks VS

Abstract

A neural network is a nonparametric statistical modeling procedure known for its capacity to process nonlinear relationships. For estimating the previous diameter of a tree, the exact functional relationship between the response variable and the independent variables is unknown. The relationship is most likely nonlinear. Multiple linear regression was used to develop a model for estimating the previous diameter of trees in West Virginia. The data were split into a model data set with 8,723 observations and a validation data set with 7,951 observations. The dependent variable was either basal-area increment or diameter increment. Two different sets of independent variables were evaluated. The data were divided into six species groups based on the rank of the average diameter growth of the species. Basal-area increment was a superior dependent variable for the multiple linear regression model. Basal-area increment is a nonlinear transformation of the diameter increment. It was thought that neural networks with its capacity to capture nonlinear relationships might provide an equivalent or superior solution with the diameter increment response variable as opposed to the basal-area increment response variable. All of the basal-area increment models had a higher R2 than their diameter increment counterparts. Neither technique was superior in all cases. Other issues such as the stopping criteria, initial weight selection, the optimal number of hidden nodes, and the optimal number of hidden layers in a neural network are also discussed.

Parent Publication

• 12th Central Hardwood Forest Conference

Citation

King, Susan L. 1999. Neural networks VS. multiple linear regression for estimating previous diameter. In: Stringer, Jeffrey W.; Loftis, David L., eds. 1999. Proceedings, 12th central hardwood forest conference; 1999 February 28-March 1-2; Lexington, KY. Gen. Tech. Rep. SRS-24. Asheville, NC: U.S. Department of Agriculture, Forest Service, Southern Research Station. 293 p