Pengembangan Framework Ensemble Hibrida Berbasis Multimodal untuk Prediksi Sinyal Transaksi Saham bagi Swing Trader

Wijaya, Arya Yudhi (2026) Pengembangan Framework Ensemble Hibrida Berbasis Multimodal untuk Prediksi Sinyal Transaksi Saham bagi Swing Trader. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7025211015-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Pergerakan harga saham yang dipengaruhi berbagai faktor menyebabkan pengambilan keputusan investasi menjadi kompleks. Investor, khususnya swing trader, membutuhkan rekomendasi transaksi berupa sinyal Buy, Hold, dan Sell, bukan hanya prediksi harga saham. Akan tetapi, sebagian besar penelitian terdahulu masih menggunakan sumber data tunggal dan model prediksi yang berdiri sendiri. Oleh karena itu, penelitian ini mengusulkan model prediksi sinyal transaksi saham berbasis data multimodal yang mengintegrasikan data teknikal dan sentimen berita melalui Framework Ensemble Hibrida.
Penelitian ini menggunakan data historis saham dan berita keuangan Indonesia periode 25 September 2018 hingga 25 September 2023. Data historis berupa data perdagangan saham dan indikator teknikal, sedangkan berita diproses melalui analisis sentimen untuk menghasilkan fitur sentimen. Kedua sumber data tersebut diintegrasikan menjadi dataset multimodal yang selanjutnya diproses melalui tiga jalur prediksi, yaitu Framework Prediksi Harga berbasis Recurrent Neural Network, Voting Ensemble Machine Learning, dan Voting Ensemble Deep Learning. Keluaran dari ketiga jalur tersebut kemudian digabungkan menggunakan mekanisme majority voting untuk menghasilkan sinyal Buy, Hold, atau Sell.
Pengujian dilakukan pada sepuluh saham LQ45 BEI dengan horizon prediksi 1, 5, 10, 20, dan 50 hari perdagangan. Hasil penelitian menunjukkan bahwa pendekatan multimodal mampu meningkatkan kualitas prediksi dibandingkan penggunaan sumber data tunggal. Framework yang diusulkan juga menghasilkan performa yang lebih stabil dibandingkan model individual, dengan akurasi rata-rata mencapai 90.53% pada horizon prediksi 50 hari. Temuan ini menunjukkan bahwa integrasi data teknikal dan sentimen berita melalui pendekatan ensemble berpotensi menjadi solusi yang efektif untuk mendukung sistem pendukung keputusan investasi.
=========================================================================================================================================
Stock price movements are influenced by various internal and external factors, making investment decision-making a complex task. Investors, particularly swing traders, require transaction recommendations in the form of Buy, Hold, and Sell signals rather than merely stock price predictions. However, most previous studies have relied on a single data source and standalone prediction models. Therefore, this study proposes a multimodal stock transaction signal prediction model that integrates technical data and news sentiment through a Hybrid Ensemble Framework.
This study utilizes historical stock data and Indonesian financial news collected from September 25, 2018, to September 25, 2023. Historical data consist of stock trading data and technical indicators, while news articles are processed using sentiment analysis to generate sentiment features. These two data sources are integrated into a multimodal dataset and subsequently processed through three prediction pathways: a Recurrent Neural Network-based Price Prediction Framework, a Machine Learning Voting Ensemble, and a Deep Learning Voting Ensemble. The outputs of these three pathways are then combined using a majority voting mechanism to generate Buy, Hold, or Sell signals.
Experiments were conducted on ten LQ45 stocks listed on the Indonesia Stock Exchange across prediction horizons of 1, 5, 10, 20, and 50 trading days. The results demonstrate that the multimodal approach improves prediction performance compared with the use of a single data source. The proposed framework also achieves more stable performance than individual models, with an average accuracy of 90.53% at the 50-day prediction horizon. These findings indicate that the integration of technical data and news sentiment through an ensemble approach has strong potential as an effective solution for supporting investment decision support systems.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Prediksi Saham, Multimodal Financial Prediction, Framework Ensemble Hibrida, Machine Learning, Deep Learning, Stock Trading Signal Prediction, Multimodal Financial Prediction, Hybrid Ensemble Framework, Machine Learning, Deep Learning
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA9.58 Algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis (Comp Science)
Depositing User:	Arya Yudhi Wijaya
Date Deposited:	29 Jul 2026 06:02
Last Modified:	29 Jul 2026 06:02
URI:	http://repository.its.ac.id/id/eprint/139205

Actions (login required)

View Item