TITLE:
Hybrid ARIMA/RBF Framework for Prediction BUX Index
AUTHORS:
Dušan Marček
KEYWORDS:
High Frequency Data, Statistical Forecasting Models, Rbf Neural Networks, Machine Learning
JOURNAL NAME:
Journal of Computer and Communications,
Vol.3 No.5,
May
25,
2015
ABSTRACT:
In this paper, we construct and
implement a new architecture and learning method of customized hybrid RBF neural
network for high frequency time series data forecasting. The hybridization is carried
out using two running approaches. In the first one, the ARCH (Autoregressive Conditionally
Heteroscedastic)-GARCH (Generalized ARCH) methodology is applied. The second modeling
approach is based on RBF (Radial Basic Function) neural network using Gaussian activation
function with cloud concept. The use of both methods is useful, because there is
no knowledge about the relationship between the inputs into the system and its output.
Both approaches are merged into one framework to predict the final forecast values.
The question arises whether non-linear methods like neural networks can help modeling
any non-linearities being inherent within the estimated statistical model. We also
test the customized version of the RBF combined with the machine learning method
based on SVM learning system. The proposed novel approach is applied to high frequency
data of the BUX stock index time series. Our results show that the proposed approach
achieves better forecast accuracy on the validation dataset than most available
techniques.