Wong, Pee Shoong (2025) Three-point moving gradient with long short-term memory for stock prices time series forecasting. Doctoral thesis, Universiti Teknikal Malaysia Melaka.
![[img]](http://eprints.utem.edu.my/style/images/fileicons/text.png) |
Text (24 Pages)
Three-point moving gradient with long short-term memory for stock prices time series forecasting (24 pages).pdf - Submitted Version Download (741kB) |
|
Text (Full Text)
Three-point moving gradient with long short-term memory for stock prices time series forecasting.pdf - Submitted Version Restricted to Registered users only Download (5MB) |
Abstract
The forecasting the stock market has always been a fascination among investors and speculators. The significance of this research is that it is to provide an alternative of forecasting time series prices technique apart from the standard techniques such as Long Short-Term Memory (LSTM). Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN) etc. The objectives of this research is to propose Three-Point Moving Gradient (TPMG) with Long Short Term Memory (LSTM) technique in stock price forecasting to have linkage between the immediate past, the present and the immediate future values in standard LSTM, and to evaluate the performance accuracy of the proposed TPMG with the standard LSTM and CNN for the actual Kuala Lumpur Stock Exchange (KLSE) stock price. This research feeds daily market stock prices into LSTM CNN and price gradients into TPMG-LSTM networks and then compare the forecasting results of each network to find out which is closest to the observed price. The main data source is the webpage https://www.malaysiastock.bizlMarket-Watch.aspx. From the forecasting testing data collected. TPMG-LSTM Residual Analysis(RA)(-0.0200) is closer to the real data than LSTM RMSE (-0.0306) or CNN RMSE (0.4480). TPMG-LSTM Coefficient of Determination G2) (-0.4782) is closer to the real data than LSTM-LSTM Rt (-4.1052) or CNN R2 (-12.4956). TPMG-LSTM Mean Absolute Error (MAE) (0.0391) is closer to the real data than LSTM MAE (0.1179) or CNN RMSE (0.4831). TPMG-LSTM Root Mean Square Error (RMSE) (0.0486) is closer to the real data than LSTM RMSE (0.1399) or CNN RMSE (0.5190). All results show that TPMGLSTM produces forecasting closest to the observed data compared to standard LSTM or CNN.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | Forecasting, LSTM, Time series, Stock price, Prediction |
| Subjects: | T Technology T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Faculty of Information and Communication Technology |
| Depositing User: | Norhairol Khalid |
| Date Deposited: | 21 Jan 2026 07:19 |
| Last Modified: | 21 Jan 2026 07:19 |
| URI: | http://eprints.utem.edu.my/id/eprint/29449 |
| Statistic Details: | View Download Statistic |
Actions (login required)
 |
View Item |