Prediksi Harga Saham Menggunakan Ensambel GRU-Attention dan Random Forest Regressor dengan Faktor Eksternal

Sjahrunnisa, Anita (2024) Prediksi Harga Saham Menggunakan Ensambel GRU-Attention dan Random Forest Regressor dengan Faktor Eksternal. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025221016-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (6MB) | Request a copy

Abstract

Perkembangan ekonomi global berdampak signifikan pada bisnis di seluruh dunia, termasuk di Indonesia, yang mengakibatkan pertumbuhan perusahaan dan meningkatnya minat masyarakat untuk berinvestasi melalui saham. Investasi saham menawarkan keuntungan yang tinggi tetapi juga risiko yang signifikan, terutama bagi investor pemula. Prediksi harga saham menjadi penting untuk membantu investor mengambil keputusan yang tepat. Dalam era teknologi saat ini, penggunaan machine learning dan deep learning untuk prediksi harga saham menjadi semakin umum. Penelitian ini bertujuan untuk memprediksi harga saham menggunakan kombinasi model GRU-Attention dan Random Forest Regressor dengan mempertimbangkan faktor eksternal seperti suku bunga, inflasi, dividen, dan hari (day of the week effect).
Beberapa penelitian sebelumnya telah menggunakan metode seperti deep learning (LSTM, CNN, dan lain-lain) serta machine learning (KNN, SVM, dan lain-lain), tetapi kebanyakan hanya diaplikasikan pada data saham dari beberapa perusahaan tanpa variasi parameter dan tanpa faktor eksternal. Dengan kondisi pasar saham yang tidak menentu dan banyaknya faktor membuat prediksi menjadi sulit. Oleh karena itu, penelitian ini menggunakan kombinasi GRU-Attention dan Random Forest Regressor dengan data harga saham historis dari beberapa perusahaan besar di Indonesia dan faktor eksternal tersebut.
Penelitian ini menggunakan data saham dari 33 perusahaan di Indonesia yang tersebar di 11 sektor saham, diambil dari periode 1 Januari hingga 12 Februari 2024. Fokus utama penelitian ini adalah: menganalisis pengaruh faktor eksternal (dividen, suku bunga, inflasi, dan hari (day of the week effect)) dalam prediksi harga saham, menyusun langkah-langkah sistematis dalam melakukan prediksi harga saham, dan mengevaluasi kinerja prediksi harga saham menggunakan ensambel GRU-Attention dan Random Forest Regressor yang dibandingkan dengan metode pembanding (GRU, Random Forest Regressor, GRU-Attention, dan GRU-Random Forest Regressor). Penggunaan data historis menghasilkan model dengan kinerja prediksi MAE 0,02025, MSE 0,00135, RMSE 0,02897, MAPE 8,36661, dan R-Squared 0,95401. Penggabungan data historis dan faktor eksternal menghasilkan model dengan kinerja prediksi MAE 0,01622, MSE 0,00060, RMSE 0,02244, MAPE 7,113875, dan R-Squared 0,97063. Ini menunjukkan bahwa model yang dihasilkan dari penggabungan fitur data historis dan faktor eksternal memberikan prediksi dengan kesalahan lebih rendah, sehingga dapat menjadi panduan yang lebih baik bagi investor dalam membuat keputusan investasi.
============================================================================================================================
The global economic developments have significantly impacted businesses worldwide, including in Indonesia, resulting in the growth of companies and increased public interest in investing in stocks. Stock investments offer high returns but also come with significant risks, especially for novice investors. Accurate stock price predictions are essential to assist investors in making informed decisions. In the current technological era, the use of machine learning and deep learning for stock price prediction has become increasingly common. This study aims to predict stock prices using a combination of GRU-Attention and Random Forest Regressor models, considering external factors such as interest rates, inflation, dividends, and days of the week.
Previous research has employed methods such as deep learning (LSTM, CNN, etc.) and machine learning (KNN, SVM, etc.), but most have been applied to stock data from a limited number of companies without varying parameters and excluding external factors. The unpredictable stock market and the numerous influencing factors make accurate predictions challenging. Therefore, this study utilizes the combination of GRU-Attention and Random Forest Regressor models with historical stock price data from several major companies in Indonesia, incorporating the mentioned external factors.
This research utilizes stock data from 33 companies in Indonesia, spanning 11 stock sectors, collected from January 1, 2015, to February 12, 2024. The main focuses of this research are: analyzing the impact of external factors (dividends, interest rates, inflation, and day of the week effect) on stock price predictions, developing systematic steps for stock price prediction, and evaluating the performance of stock price predictions using the GRU-Attention ensemble and Random Forest Regressor, compared with benchmark methods (GRU, Random Forest Regressor, GRU-Attention, and GRU-Random Forest Regressor). Using historical data, the model achieved prediction performance with an MAE of 0.02025, MSE of 0.00135, RMSE of 0.02897, MAPE of 8.36661, and R-Squared of 0.95401. The combination of historical data and external factors improved the model's performance, achieving an MAE of 0.01622, MSE of 0.00060, RMSE of 0.02244, MAPE of 7.113875, and R-Squared of 0.97063. These results indicate that the model combining historical data and external factors provides more accurate predictions with lower errors, thereby offering better guidance for investors in making investment decisions.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	GRU, Attention, ensambel, Random Forest Regressor, saham, suku bunga, inflasi, dividen, hari, GRU, Attention, ensemble, Random Forest Regressor, stocks, interest rates, inflation, dividends, day of the week.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Anita Sjahrunnisa
Date Deposited:	06 Aug 2024 08:53
Last Modified:	06 Aug 2024 08:53
URI:	http://repository.its.ac.id/id/eprint/111345

Actions (login required)

View Item