Homogeneous ensemble Neural Cognition for CIMB stock price prediction
Stock market forecasting has always been a topic of research interest due to the lucrative profit. However, stock market forecasting is a complicated task because of the not linear, volatile and random nature of the stock market. Literature reviews have shown that Artificial Neural Network (ANN) is...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | https://eprints.ums.edu.my/id/eprint/26958/1/Homogeneous%20Ensemble%20Neural%20Cognition%20for%20CIMB%20Stock%20Price%20Prediction.pdf https://eprints.ums.edu.my/id/eprint/26958/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Stock market forecasting has always been a topic of research interest due to the lucrative profit. However, stock market forecasting is a complicated task because of the not linear, volatile and random nature of the stock market. Literature reviews have shown that Artificial Neural Network (ANN) is an appropriate technique to be used in forecasting activities. However, there are certain limitation in single neural network. Therefore ensemble tecniques is introduced to overcome the limitation of single neural network. Ensemble techniques overcome the limitation of single learner by covering the different area of the problem search space aggregating multiple learner, consequently, reducing the error of estimation. Hence, this research investigates the performance of homogeneous ensemble neural network in closing price prediction. Ensemble Neural Network (ENN) has shown to be able to overcome the limitation of single neural network by covering the different area of the problem search space aggregating multiple single neural network, consequently, reducing the error of estimation. As the thesis title implies, the investigation will be focusing on homogeneous ensemble neural network which means multiple same architecture neural network will be used in the ensemble neural network. The two single neural network architectures as the building block for homogeneous ENN are used in this thesis which are FeedForward Neural Network (FFNN) and Recurrent Neural Network (RNN). These two architectures are among the ANN architectures that are widely adopted in ANN research. As shown in the literature review, both architectures shown to perform well in different studies which using different parameters and network configuration that have been conducted. Hence, it is also the interest of this thesis to the performance comparison of FFNN and RNN as well in this case study. The resulted single FFNN and RNN learner conducted in the experiment will then be used in experiment of homogeneous ENN. The results of these experiments are compared against each other to see how well the homogeneous ENN can perform better than single FFNN and RNN. The study case adopted in this thesis is CIMB stock listed in KLSE. CIMB is a well established financial bank company in Malaysia. The stock is selected due the volatile pattern trend that is suitable as the study case. In this study, a collection of input parameters which include technical data and economic variables are presented to the ANN to forecast the stock closing price of CIMB. The performance of the different ANN architectures are empirically evaluated based on Mean Square Error (MSE) and prediction accuracy. |
|---|