Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

 

A STOCK PRICE PREDICTION MODEL BY THE NEURAL NETWORK APPROACH

 

TITLE A STOCK PRICE PREDICTION MODEL BY THE NEURAL NETWORK APPROACH.
AUTHOR PRAPAPHAN PAN-O
DEGREE MASTER OF SCIENCE PROGRAMME IN COMPUTER SCIENCE
FACULTY FACULTY OF SCIENCE
ADVISOR SUPACHAI TANGWONGSAN
CO-ADVISOR SUDSANGUAN NGAMSURIYAROJ
 
ABSTRACT
This research proposes a stock price prediction based on the backpropagation neural network approach. The focus of this scheme is to adjust the neural network training methodology for improving the accuracy of both value prediction and value fluctuation direction of the desired value. The predicted results of the traditional learning methodology and those derived from using the approach will be compared. The process used to train the network is to feed a large number of inputs to the network in order to reach the minimum mean squared error between the desired value and the actual output in each training cycle. This process may give the predicted value close to the desired value. We proposed the method to adapt the training methodology using the Two-Step Continuous Fluctuation Direction (TSCFD) of the desired value for calculating the changing weight. To perform the prediction task, the proposed system is tested using the time series data from 3 sources: namely, 2 closing stock prices with 500 trading days and a generated data series with 487 observations from Mackey-Glass Equation. The general prediction measurements such as Mean Squared Error (MSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) are used to determine how close the desired value is. Additionally, we use the Prediction Of Correct Fluctuation Direction (POCFD) to indicate the correctness of the predicted fluctuation direction while Tolerance1% and Tolerance5% are used to determine how close it is for the specified range of desired value. The model performance is measured with various settings of network parameters and topologies. The average of experimental results derived from using the TSCFD for predicting the next value point only by using 20-2-1 network can reach 75%, 90% and 80% accuracy by the Tolerance1%, Tolerance5% and POCFD, respectively. In conclusion, it was found that the experiments yielded quite satisfactory results and the research objectives were achieved.
KEYWORD NEURAL NETWORK / BACKPROPAGATION / PREDICTION / STOCK PRICE / DIRECTION

 

Go to Top

 

ICT Building, Mahidol University, 999 Phuttamonthon 4 Road, Salaya, Nakhonpathom 73170 Tel. +66 02 441-0909 Fax. +66 02 849-6099
Mahidol University Computing Center, The Faculty of ICT, Mahidol University , Rama 6 Road, Rajathevi, Bangkok 10400 Tel. +66 02 354-4333 Fax. +66 02 354-7333