Peramalan Harga Saham Adro Menggunakan Pendekatan Hibrida Roberta-Bilstm Berbasis Sentimen Pasar Dan Harga Historis

Farhan, Farhan Muhammad Rizqi (2026) Peramalan Harga Saham Adro Menggunakan Pendekatan Hibrida Roberta-Bilstm Berbasis Sentimen Pasar Dan Harga Historis. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221035-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Volatilitas harga saham pada pasar negara berkembang semakin dipengaruhi oleh faktor nonlinier yang tidak sepenuhnya tercermin dalam data historis, khususnya sentimen pasar yang terbentuk dari pemberitaan media. Kondisi tersebut mendorong pengembangan pendekatan peramalan yang mampu mengintegrasikan data kuantitatif dan kualitatif secara simultan. Penelitian ini bertujuan untuk mengembangkan dan mengevaluasi model hibrida RoBERTa–Bidirectional Long Short-Term Memory (BiLSTM) dalam memprediksi harga saham PT Adaro Energy Indonesia Tbk (ADRO). Data penelitian mencakup harga penutupan saham, volume perdagangan, serta sentimen berita yang diperoleh dari portal Detik.com dan CNBC Indonesia pada periode Januari 2024 hingga Juli 2025. Ekstraksi sentimen dilakukan menggunakan model w11wo/indonesian-roberta-base-sentiment-classifier untuk menghasilkan skor sentimen harian. Seluruh variabel dinormalisasi dan dimodelkan menggunakan arsitektur BiLSTM dengan skema prediksi satu hari ke depan. Evaluasi kinerja model dilakukan menggunakan metrik Mean Absolute Percentage Error (MAPE) dan Root Mean Square Error (RMSE). Hasil penelitian menunjukkan bahwa integrasi sentimen berita ke dalam model BiLSTM menghasilkan peningkatan akurasi prediksi yang signifikan dibandingkan model tanpa sentimen. Model hibrida mencatatkan nilai MAPE sebesar 5,60% dan RMSE sebesar 149,54, lebih unggul dibandingkan model baseline tanpa sentimen yang memiliki nilai MAPE sebesar 13,53%. Pola prediksi model mampu mengikuti pergerakan harga aktual secara presisi, terutama pada periode volatilitas tinggi. Temuan ini menunjukkan bahwa sentimen berita memiliki kontribusi signifikan dalam meningkatkan akurasi peramalan harga saham ADRO serta menegaskan efektivitas pendekatan hibrida RoBERTa–BiLSTM dalam analisis pasar saham berbasis teks.
==================================================================================================================================
Stock price volatility in emerging markets is increasingly influenced by nonlinear factors that are not fully captured by historical price data, particularly market sentiment derived from news coverage. This condition motivates the development of forecasting approaches capable of integrating quantitative and qualitative information simultaneously. This study aims to develop and evaluate a hybrid RoBERTa–Bidirectional Long Short-Term Memory (BiLSTM) model for predicting the stock price of PT Adaro Energy Indonesia Tbk (ADRO). The dataset consists of closing stock prices, trading volume, and news sentiment collected from Detik.com and CNBC Indonesia during the period from January 2024 to July 2025. Sentiment extraction is performed using the w11wo/indonesian-roberta-base-sentiment-classifier model to generate daily sentiment scores. All variables are normalized and modeled using a BiLSTM architecture with a one-day-ahead forecasting scheme. Model performance is evaluated using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). The results demonstrate that incorporating news sentiment into the BiLSTM model significantly improves forecasting accuracy compared to the model without sentiment. The hybrid model achieved a MAPE of 5.60% and an RMSE of 149.54, outperforming the baseline model, which had a MAPE of 13.53%. The predicted values closely follow actual price movements, particularly during periods of high volatility. These findings indicate that news sentiment provides a meaningful contribution to stock price forecasting and confirm the effectiveness of the RoBERTa–BiLSTM hybrid approach for text-based financial market analysis.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Peramalan Harga Saham, Analisis Sentimen, RoBERTa, Bidirectional LSTM, Berita Keuangan. ADRO, Behavioral Finance, BiLSTM, Hybrid Model, RoBERTa, Sentiment Analysis, Stock Price Forecasting.
Subjects:	Q Science Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry)
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Farhan Muhammad Rizqi
Date Deposited:	04 Aug 2026 07:31
Last Modified:	04 Aug 2026 07:31
URI:	http://repository.its.ac.id/id/eprint/140535

Actions (login required)

View Item