Repozytorium publikacji - Politechnika Gdańska

Stock market is a complex and dynamic industry that has always presented challenges for stakeholders and investors due to its unpredictable nature. This unpredictability motivates the need for more accurate prediction models. Traditional prediction models have limitations in handling the dynamic nature of the stock market. Additionally, previous methods have used less relevant data, leading to suboptimal performance. This study proposes the use of Bidirectional Encoder Representations from Transformers (BERT), a pre-trained Large Language Model (LLM), to predict Dhaka Stock Exchange (DSE) market movements. We also introduce a new dataset designed specifically for this problem, capturing important characteristics and patterns that were missing in other datasets. We test our new dataset of headlines and stock market indexes on various machine learning techniques, including Decision Tree (DT), Logistic Regression (LR), K-Nearest Neighbors (KNN), Random Forest (RF), Linear Support Vector Machine (LSVM), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs), Bidirectional Long Short-Term Memory (Bi-LSTM), BERT, Financial Bidirectional Encoder Representations from Transformers (FinBERT), and RoBERTa, which are compared to assess their predictive capabilities. Our proposed model achieves 99.83% accuracy on the training set and 99.78% accuracy on the test set, outperforming previous methods.

Autorzy

• Md. Nabil Rahman Khan,
• Md Rafiqul Islam,
• Cesar Sanin,
• prof. dr hab. inż. Edward Szczerbicki link otwiera się w nowej karcie