<![CDATA[Stock price volatility in the Food and Beverage (F&B) sector presents persistent challenges for investors and decision-makers, particularly in emerging markets. This study proposes an automated stock price prediction framework whose primary contribution lies in the system-level integration of a Stacked Long Short-Term Memory (LSTM) model with a scalable big data orchestration pipeline, rather than in introducing a new forecasting algorithm alone. The system targets three Indonesian F&B companies PT Indofood CBP Sukses Makmur Tbk, PT Mayora Indah Tbk, and PT Garudafood Putra Putri Jaya Tbk using historical daily stock price data. The dataset spans multiple years of trading records retrieved from the Yahoo Finance API, and predictions are generated for a seven-day forecasting horizon. Methodologically, the approach combines a multi-layer LSTM architecture with Apache Spark for distributed data preprocessing, Apache Airflow for automated workflow orchestration, and PostgreSQL for structured data storage. This integration enables scheduled data ingestion, reproducible model training, and continuous forecasting within an end-to-end analytics pipeline. Model performance is evaluated using error-based metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE), and is benchmarked against a conventional single-layer LSTM without pipeline orchestration. Empirical results show that the proposed pipeline-based Stacked LSTM achieves lower prediction error, with MAPE values ranging between approximately 1.1% and 2.2% across the evaluated stocks, indicating improved stability and accuracy. Overall, the findings demonstrate enhanced forecasting reliability and deployment readiness through automated pipelines.]]>
Copyrights © 2026
