Data & Forecasting · 2024
Generative AI Saham
Market forecasting informed by news sentiment
A market-forecasting application that connects five years of price history with news sentiment and LSTM time-series modeling.
- LSTM
- TensorFlow
- Keras
- Dash
- Plotly
- yfinance
Measured evidence
- BBNI R² score
- 0.9733
- BBNI RMSE
- 61.74
- historical price data
- 5 years
Problem worth solving
Explore whether financial-news sentiment can add useful context to a stock-price forecasting pipeline.
Engineering decisions
- Automated yfinance ingestion, Detik.com news scraping, sentiment mapping, MinMaxScaler normalization, and 60-step windows.
- Trained a sequential LSTM with sentiment-feature concatenation.
- Built a Dash and Plotly interface for experiment selection, stock exploration, and model evaluation.
What changed
- Evaluated five input scenarios with RMSE, MAE, and R².
- The reported BBNI run using Siebert news sentiment reached RMSE 61.74 and R² 0.9733.