Forecasting Performance of Cascade Backward Neural Network for Data with Outliers

Abstract

In this study, a cascade backward propagation neural network (CBPNN) technique was
suggested with the focus of determining its performance over dataset that are presumed to
contain outliers which inturns leads to a heteroscedastic relationship. The new developed
technique was tested using the Aboline, Airfoil, Istanbul stock exchange, Earthquake and
Attenuation dataset obtained from the UCI machine learning repository and R dataset
website. The performance of the new technique was compared with the performances of
the weighted least square (WLS) technique. The comparison was carried out via the
evaluating metrics of the mean square error (MSE), root mean square error (RMSE), mean
absolute error (MAE) as well as the mean absolute percentage error (MAPE). The
comparison indicates that the results emerging from the developed (CBPNN) technique
gives a better performance when compared with the WLS regression technique for all the
dataset considered via the evaluating metrics used. Analysis for this study was carried out
using MATLAB R2014a and R i386 version 3.3.0 software’s and the results are presented.
The regression technique was done using the R software while the neural network aspect
was 5done using the MATLAB R2014a software.