Comparison Artificial Neural Network and Linear Regression Methods

Select Volume / Issue:

• Volume 2,Issue 1, Jan. 2014

Year:

2014

Type of Publication:

Article

Keywords:

Artificial Neural Network, Multi-Layer Feed Forward Network, Data Mining, Linear Regression, Statistical Method

Authors:

N. Abbasi; A. Falamaki; F. Afshar

Journal:

IJISM

Volume:

2

Number:

1

Pages:

14-22

Month:

January

Abstract:

In this paper, implementation of statistical methods and data mining is compared by means of changing the number and type of variables and increasing the sample size. Moreover, effect of increment hidden layers, increment the number neurons of network, increment the number of independent and classified variables and increment the sample size is studied. Hence data mining and statistical methods applied on five simulated examples with artificial neural network and linear regression in order to compare these methods. SSE (sum of squares error) values used and some additional findings yielded: 1- When independent variables are continuous and classified even by varying the number of variables and sample size, neural networks have better performance. 2-It couldn't indicate certainly whether increment hidden layers of network and the number of neurons in them would improve its implementation or not. Only by repeating and applying trial and error could achieve best network geometry for finding optimal solution. 3- Increase in the number of variables and existence of classified variables in model would reduce quality of performance of neural network. 4- By having classified variables, increment in the number of hidden layers of network leads to increase in SSE values.

Full text: IJISM_09_Final.pdf [Bibtex]