Why
Stock Price Prediction Remains a Challenge
Stock
prices are shaped by many interacting factors, including company fundamentals,
macroeconomic conditions, investor sentiment, and unexpected external shocks.
As a result, price movements are often volatile and difficult to anticipate.
Traditional forecasting methods, such as moving averages or linear statistical
models, tend to struggle with these nonlinear and rapidly changing patterns.
The
Gated Recurrent Unit (GRU) is a streamlined version of this technology.
It was developed to solve common problems in older neural networks, especially
the loss of important information over time. By using a gating mechanism, GRU
models can decide which past information should be retained and which can be
ignored. This makes them well suited for financial time series, where past
trends often influence future prices.
How
the Study Was Conducted
The research team from Universitas Terbuka Medan analyzed historical closing prices of Bank Syariah Indonesia shares obtained from a public financial data platform. To ensure fairness and consistency, the dataset was divided into two parts:
- 80 percent of the data was used to train the model.
- 20 percent was reserved for testing its predictive accuracy.
Before
analysis, the data was standardized so that large price differences would not
distort the learning process. This step helps AI models recognize patterns more
reliably.
Key
Findings: Simpler Model Performs Best
One
of the most striking results of the study is that the simplest model
delivered the best performance. The baseline GRU model consistently
produced lower prediction errors than the more complex four-layer and
five-layer versions.
- Mean Squared Error (MSE): 5,601.44
- Mean Absolute Error (MAE): 54.06
- Root Mean Squared Error (RMSE): 74.84
- Mean Absolute Percentage Error (MAPE): 1.90%
In
comparison, the four-layer and five-layer models showed higher error rates
across all metrics. Adding more layers increased computational complexity but
did not improve accuracy. In fact, deeper models tended to overfit the data,
making them less reliable for prediction.
Real-World
Implications
The findings have several practical implications:
- For investors, GRU-based predictions can complement traditional analysis and help identify short-term trends.
- For financial institutions, the approach offers a foundation for developing internal decision-support systems powered by artificial intelligence.
- For researchers and educators, the study highlights the importance of model selection and shows that simpler architectures may be more effective in certain contexts.
- For policymakers, improved forecasting tools can support market monitoring and financial stability efforts, particularly in the rapidly expanding sharia banking sector.
By
demonstrating that a baseline GRU model performs best, the research also
encourages more efficient use of computational resources. This is especially
relevant for institutions with limited access to high-end computing
infrastructure.
Author Profiles
Muhammad Nurmarendra, S.Kom. – Lecturer and researcher at Universitas Terbuka Medan, Specializing in machine learning, data analytics, and intelligent systems.
Anik,
S.Kom., M.Kom.
– Academic staff at Universitas Terbuka Medan
Expertise in data
processing and artificial intelligence.
Yasir Riady, S.T., M.T. – Researcher in information technology and computational modeling at Universitas Terbuka Medan.
Source
Muhammad Nurmarendra, Anik, Yasir
Riady, “Predicting the Share Price of Bank Syariah Indonesia Using the GRU
(Gated Recurrent Unit) Algorithm". Asian Journal of Management Analytics (AJMA) Vol. 5, No. 1 2026:
43-56
DOI:https://doi.org/10.55927/ajma.v5i1.15964
URL : https://journal.formosapublisher.org/index.php/ajma
 Algorithm){kind=link}
0 Komentar