ARIF, Erman and Suherman, Suherman and Widodo, Aris Puji (2026) INTEGRASI DATA SENTIMEN MEDIA SOSIAL DAN DATA HISTORIS SAHAM UNTUK MENINGKATKAN AKURASI PREDIKSI HARGA SAHAM BANK DIGITAL MENGGUNAKAN ALGORITMA MACHINE LEARNING. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text cover aja.pdf Download (236kB)
	Text Bagian Awal.pdf Restricted to Repository staff only Download (675kB)
	Text BAB 1.pdf Download (311kB)
	Text BAB 2.pdf Download (560kB)
	Text BAB 3.pdf Restricted to Repository staff only Download (439kB)
	Text BAB 4.pdf Restricted to Repository staff only Download (661kB)
	Text BAB 5.pdf Restricted to Repository staff only Download (282kB)
	Text Daftar Pustaka.pdf Download (207kB)
	Text Lampiran.pdf Restricted to Repository staff only Download (1MB)

Abstract

Penelitian ini mengkaji korelasi integrasi data historis harga saham dan sentimen media sosial terhadap fluktuasi harga saham PT Bank Jago Tbk sebagai representasi bank digital di Indonesia, sekaligus mengevaluasi dampaknya terhadap akurasi prediksi. Latar belakang penelitian didorong oleh pesatnya pertumbuhan perbankan digital dan dominasi investor ritel yang aktif di media sosial, namun bukti empiris lokal terkait efektivitas integrasi sentimen masih terbatas. Penelitian ini menawarkan kebaruan berupa rancangan alur kerja integrasi data yang mencakup pembersihan teks, pelabelan sentimen menggunakan IndoBERT, penyelarasan temporal dengan kalender bursa, serta rekayasa fitur gabungan (indikator teknikal dan skor sentimen). Model prediksi dibangun menggunakan tiga algoritma yang mewakili pendekatan linier, non-linier, dan deep learning, yaitu Regresi Linear, Random Forest, dan Neural Network, dengan evaluasi berbasis metrik MAE, RMSE, dan R² serta validasi silang K-Fold.
Hasil penelitian menunjukkan bahwa integrasi sentimen media sosial tidak memberikan peningkatan akurasi yang signifikan dibandingkan model berbasis data historis saja. Selisih MAE dan RMSE antar model sangat kecil, sementara nilai R² keduanya mendekati 1,00, menegaskan dominasi data historis sebagai prediktor utama. Analisis korelasi melalui scatter plot dan distribusi residual mengonfirmasi bahwa hubungan antara skor sentimen dan perubahan harga saham bersifat lemah. Faktor penghambat utama meliputi kualitas data sentimen, asinkronisasi waktu, dan keberadaan ironi/sarkasme. Temuan ini mendukung hipotesis pasar efisien bentuk lemah (EMH) dan memberikan perspektif behavioral finance terkait keterbatasan pengaruh sentimen publik dalam horizon pendek.
Implikasi penelitian ini bersifat metodologis dan praktis. Alur kerja integrasi data yang dirumuskan dapat menjadi acuan untuk studi lanjutan, sementara hasil empiris menegaskan bahwa sentimen sebaiknya diposisikan sebagai indikator pelengkap, bukan variabel utama, dalam sistem prediksi harga saham. Untuk meningkatkan relevansi, penelitian mendatang disarankan mengadopsi teknik NLP yang lebih canggih, memperluas sumber sentimen, dan mengintegrasikan data secara real-time agar mampu menangkap dinamika pasar secara lebih akurat.
Kata Kunci: Korelasi Harga Saham, Sentimen Media Sosial, Bank Digital, Machine Learning, Regresi Linear, Random Forest, Neural Network

This study examines the correlation between the integration of historical stock price data and social media sentiment with stock price fluctuations of PT Bank Jago Tbk, representing digital banking in Indonesia, while also evaluating its impact on prediction accuracy. The research background is driven by the rapid growth of digital banking and the dominance of retail investors active on social media, yet local empirical evidence regarding the effectiveness of sentiment integration remains limited.
The novelty of this study lies in the design of a data integration workflow that includes text cleaning, sentiment labeling using IndoBERT, temporal alignment with the stock exchange calendar, and combined feature engineering (technical indicators and sentiment scores). The prediction models are built using three algorithms representing linear, non-linear, and deep learning approaches: Linear Regression, Random Forest, and Neural Network, with evaluation based on MAE, RMSE, and R² metrics and K-Fold cross-validation.
The findings indicate that integrating social media sentiment does not significantly improve accuracy compared to models based solely on historical data. Differences in MAE and RMSE across models are minimal, while R² values for both approaches are close to 1.00, reaffirming the dominance of historical data as the primary predictor. Correlation analysis through scatter plots and residual distribution confirms that the relationship between sentiment scores and stock price changes is weak. Key limiting factors include sentiment data quality, temporal asynchrony, and the presence of irony/sarcasm. These findings support the weak- form Efficient Market Hypothesis (EMH) and provide a behavioral finance perspective on the limited influence of public sentiment in the short term.
The implications of this research are both methodological and practical. The proposed data integration workflow can serve as a reference for future studies, while empirical results suggest that sentiment should be positioned as a complementary indicator rather than a primary variable in stock price prediction systems. To enhance relevance, future research is recommended to adopt more advanced NLP techniques, expand sentiment sources, and integrate data in real time to better capture market dynamics.
Keywords: Correlation, Stock Price Prediction, Social Media Sentiment, Digital Banking, Machine Learning, Linear Regression, Random Forest, Neural Network

Actions (login required)

View Item