<![CDATA[The high volatility of banking sector stocks (BBCA, BBRI, BMRI) and the limitations of conventional forecasting methods in handling non-linear data necessitate robust and adaptive predictive models. This study aims to develop an integrated stock price prediction system utilizing a Stacked Long Short-Term Memory (LSTM) architecture embedded within a Flask-based interactive web dashboard. Adopting the CRISP-DM framework, the model was trained using daily and hourly historical data from Yahoo Finance to accommodate both short-term and medium-term forecasting. Backtesting evaluation demonstrated that the LSTM model achieved Mean Absolute Percentage Error (MAPE) values below 2% for daily single-step predictions and below 0.5% for hourly intraday predictions. Furthermore, in a 7-period recursive projection, the proposed LSTM proved highly robust in mitigating error accumulation compared to Linear Regression and Support Vector Regression (SVR), successfully maintaining MAPE values below 5% for all issuers. The implementation of this dashboard system provides a significant impact on financial informatics by bridging advanced deep learning predictive algorithms into a practical, real-time decision support system for investment analysis.]]>
Copyrights © 2026
