The relationship between data skewness and accuracy of Artificial Neural Network predictive model

8th Annual International Conference 2018 on Science and Engineering

工业技术（总论）

Larasati, A.^1 ; Hajji, A.M.^2 ; Dwiastuti, Anik^1

Department of Industrial Engineering, Universitas Negeri Malang, Indonesia^1

Department of Civil Engineering, Universitas Negeri Malang, Indonesia^2

关键词: Data skewness;    Hidden layers;    Input variables;    Output variables;    Predictive modeling;    Skewed data;    Three categories;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/523/1/012070/pdf
DOI  :  10.1088/1757-899X/523/1/012070

学科分类：工业工程学

来源: IOP

【 摘 要 】

The purpose of this study is to investigate the relationship between data skewness in the output variable and the accuracy of artificial neural network predictive model. The artificial neural network predictive model is built using multilayer perceptron and consist of one output variable and six input variable, and the algorithm used is back propagation. Data used in this study is generated by conducting the simulations in 1000 cycles. Three categories of skewness used in the output variables are positive skewness, neutral, and negative skewness. The results show that data skewness does not have a significant effect on the accuracy of the artificial neural network predictive model. These results imply that artificial neural network predictive model has a higher capability to cope with skewed data due to its complexity in the hidden layer.