Peramalan Harga Saham Menggunakan Metode Hybrid Bidirectional Gated Recurrent Unit Dan Attention Mechanism

Fareliansyah, Raihan (2026) Peramalan Harga Saham Menggunakan Metode Hybrid Bidirectional Gated Recurrent Unit Dan Attention Mechanism. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221160-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Pasar modal merupakan instrumen penting dalam mendukung pergerakan sektor ekonomi, dengan sektor perbankan sebagai salah satu sektor yang memiliki fundamental kuat, stabilitas tinggi, serta prospek keberlanjutan jangka panjang. Salah satu perusahaan yang mencerminkan karakteristik tersebut adalah Bank Central Asia Tbk. (BBCA), yang tergolong sebagai saham blue chip dengan kapitalisasi pasar besar dan tingkat likuiditas yang tinggi. Permasalahan yang sering dihadapi oleh investor dan analis adalah pergerakan harga saham yang bersifat nonlinier serta dipengaruhi oleh berbagai faktor makroekonomi, seperti kondisi pemerintahan, kebijakan politik, dan perubahan indeks pasar yang direpresentasikan oleh Indeks Harga Saham Gabungan (IHSG). Untuk mengatasi permasalahan tersebut, penelitian ini mengusulkan model Bidirectional Gated Recurrent Unit (BiGRU) yang dipadukan dengan Attention Mechanism guna meningkatkan kemampuan model dalam menangkap pola temporal dan informasi penting pada data. Data yang digunakan bersifat multivariat, menggabungkan harga saham BBCA dengan indikator makroekonomi IHSG, sedangkan kinerja model dievaluasi menggunakan metrik Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), R², Directional Accuracy (DA), F1-Score, serta waktu komputasi (runtime). Berdasarkan hasil eksperimen, model BiGRU memberikan hasil terbaik dengan nilai MAPE sebesar 1,075045% menggunakan parameter 128 GRU units, learning rate 0,001, batch size 64, dan proporsi data latih dan uji sebesar 70:30. Model tersebut memerlukan waktu komputasi selama 59,64936 detik. Sementara itu, model BiGRU dengan Attention Mechanism (BiGRU-AM) menghasilkan nilai MAPE sebesar 1,075301% menggunakan 128 GRU units, learning rate 0,005, batch size 32, dan proporsi data yang sama, dengan waktu komputasi yang lebih cepat, yaitu 49,50516 detik. Meskipun model BiGRU memberikan akurasi prediksi terbaik pada data utama, model BiGRU-AM menunjukkan kemampuan generalisasi yang lebih baik pada data saham dari sektor perbankan dan industri lainnya, dengan nilai MAPE sebesar 1,6303625% pada BBRI dan 1,7195365% pada TLKM. Sebagai perbandingan, model BiGRU menghasilkan nilai MAPE sebesar 1,6584567% pada BBRI dan 1,7595084% pada TLKM, sehingga menunjukkan bahwa BiGRU-AM lebih konsisten dalam melakukan prediksi pada data yang belum pernah dilihat sebelumnya.
================================================================================================================================
The capital market is an important instrument for supporting economic growth, with the banking sector being one of the most attractive investment sectors due to its strong fundamentals, high stability, and long-term resilience. One company that exemplifies these characteristics is Bank Central Asia Tbk. (BBCA), which is classified as a blue-chip stock with large market capitalization and high liquidity. However, stock price movements are inherently nonlinear and are influenced by various macroeconomic factors, including government conditions, political policies, and fluctuations in market indices such as the Indonesian Composite Index (IHSG), making accurate forecasting a challenging task for investors and analysts. To address these challenges, this study proposes a Bidirectional Gated Recurrent Unit (BiGRU) model enhanced with an Attention Mechanism to improve the model's ability to capture temporal patterns and focus on the most relevant information in the input data. The proposed approach utilizes multivariate data consisting of BBCA stock prices combined with IHSG as a representation of macroeconomic conditions. Model performance is evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), R², Directional Accuracy (DA), F1-Score, and computational runtime. Experimental results show that the BiGRU model achieved the best prediction accuracy with a MAPE of 1.075045% using 128 GRU units, a learning rate of 0.001, a batch size of 64, and a 70:30 train-test split, requiring 59.64936 seconds of computation time. Meanwhile, the BiGRU with Attention Mechanism (BiGRU-AM) produced a slightly higher MAPE of 1.075301% using 128 GRU units, a learning rate of 0.005, a batch size of 64, and the same data split ratio, while reducing computation time to 49.50516 seconds. Although the BiGRU model achieved the highest prediction accuracy, the BiGRU-AM model demonstrated superior computational efficiency. Furthermore, in the generalization test using data from different banking and industrial sector stocks, the BiGRU-AM model achieved better performance, obtaining MAPE values of 1.6303625% for BBRI and 1.7195365% for TLKM, compared with the BiGRU model, which produced MAPE values of 1.6584567% for BBRI and 1.7595084% for TLKM. These results indicate that while the BiGRU model excels in prediction accuracy for the primary dataset, the BiGRU-AM model provides better generalization capability and computational efficiency across different stock datasets.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Attention Mechanism, BiGRU, Multivariat, Peramalan, Saham, Forecasting, Multivariate, Stocks
Subjects:	H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Raihan Fareliansyah
Date Deposited:	20 Jul 2026 07:04
Last Modified:	23 Jul 2026 03:34
URI:	http://repository.its.ac.id/id/eprint/135686

Actions (login required)

View Item