Fadhilah, Rezkihana Nur (2026) Peramalan Harga Saham di Indeks IDX ESG Leaders Menggunakan Model Fusi Data Multimodal Berbasis GRU dengan Attention Mechanism. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2043221116-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2043221116-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy |
Abstract
Investasi berbasis Environment, Social, and Governance (ESG) yang telah menjadi arus utama menciptakan risiko information overload, di mana investor kesulitan mengukur dampak gabungan antara data finansial dan sentimen pasar secara real-time. Penelitian ini bertujuan merancang model peramalan harga saham multimodal pada Indeks IDX ESG Leaders menggunakan arsitektur Gated Recurrent Unit (GRU) fusi data multimodal untuk menjembatani kesenjangan tersebut. Model yang diusulkan mengintegrasikan data kuantitatif (harga historis dan indikator teknikal) dengan data kualitatif (sentimen berita dan opini investor) melalui pendekatan intermediate fusion. Pengolahan data tekstual dilakukan menggunakan IndoSBERT untuk pemetaan relevansi semantik dan IndoRoBERTa untuk ekstraksi skor sentimen. Model dievaluasi menggunakan skema walk-forward validation dengan metrik Mean Absolute Percentage Error (MAPE) dan Root Mean Square Error (RMSE) pada horizon prediksi tiga hari ke depan. Hasil penelitian menunjukkan empat temuan utama. Pertama, kecenderungan dan intensitas sentimen negatif terbukti memiliki pengaruh yang lebih kuat terhadap pergerakan harga dibandingkan sentimen positif pada emiten IDX ESG Leaders. Kedua, integrasi data melalui model fusi multimodal GRU menghasilkan kinerja peramalan yang stabil pada horizon jangka pendek dengan tingkat akurasi tinggi pada mayoritas emiten, kecuali pada kondisi anomali ekonomi yang ekstrem. Ketiga, indikator teknikal seperti harga penutupan, Relative Strength Index (RSI), dan Average True Range (ATR) merupakan faktor pendorong dominan dalam hasil peramalan, sementara fitur sentimen berfungsi sebagai penguat sinyal pada kondisi pasar tertentu. Keempat, seluruh hasil penelitian berhasil diimplementasikan ke dalam prototipe dashboard interaktif berbasis Streamlit untuk mendukung pengambilan keputusan investasi secara transparan dan berbasis data.
===================================================================================================================================
Environment, Social, and Governance (ESG)-based investments, which have become mainstream, create the risk of information overload, where investors find it difficult to measure the combined impact of financial data and market sentiment in real time. This study aims to design a multimodal stock price forecasting model on the IDX ESG Leaders Index using the Gated Recurrent Unit (GRU) multimodal data fusion architecture to bridge this gap. The proposed model integrates quantitative data (historical prices and technical indicators) with qualitative data (news sentiment and investor opinions) through an intermediate fusion approach. Textual data processing is performed using IndoSBERT for semantic relevance mapping and IndoRoBERTa for sentiment score extraction. The model is evaluated using a walk-forward validation scheme with Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) metrics on a three-day prediction horizon. The results of the study show four main findings. First, the tendency and intensity of negative sentiment proved to have a stronger influence on price movements than positive sentiment in IDX ESG Leaders issuers. Second, data integration through the GRU multimodal fusion model produced stable forecasting performance in the short term with a high level of accuracy in the majority of issuers, except in conditions of extreme economic anomalies. Third, technical indicators such as closing price, Relative Strength Index (RSI), and Average True Range (ATR) are dominant drivers in the forecasting results, while sentiment features serve as signal amplifiers in certain market conditions. Fourth, all research results were successfully implemented into a Streamlit-based interactive dashboard prototype to support transparent and data-driven investment decision-making.
Actions (login required)
 |
View Item |