Paper
Comparative Performance of Feed Forward Neural Networks and Multiple Regression Using Simulation
-
Authors:
-
Usha A. Kumar; Mukta Paliwal
-
Abstract
-
Neural network models seem to have potential for prediction purposes and this has led to a number of studies comparing the performance of neural networks and regression analysis. Regression technique is based on certain basic assumptions and validity of these assumptions is critical to its performance. This issue does not appear to have been considered in most of the comparative studies. In the present study, we intend to focus on this aspect by comparing the performance of both the techniques using simulation when all the assumptions of regression are met. This study reveals that the performance of regression analysis and neural network are comparable for large and medium sample sizes and suggests the need for careful implementation of neural network when the sample size is small.
-
Keywords
-
Assumption Validity; Evaluation Criterion; Levenberg-Marquardt Training Algorithm; Prediction Intervals; Simulation
-
StartPage
-
167
-
EndPage
-
175
-
Doi
-