A Lambda Lakehouse Deep Learning Framework for Gold Price Forecasting in Financial Markets

Abstract

Gold remains a critical financial asset because of its dual function as both a safe-haven instrument and a key indicator of market stability. Accurate forecasting of gold prices is therefore essential for investors, policymakers, and financial institutions. This study introduces a Lambda–Lakehouse architecture integrated with deep learning models to improve the prediction accuracy of gold price time series. Historical data from 2004 to 2025 were collected, preprocessed, and managed within a cloud-based environment combining AWS S3, Apache Spark, Delta Lake, and Databricks. Three predictive models Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Transformer were implemented and evaluated using standard metrics (RMSE, MAE, MAPE, R²). Experimental results reveal that LSTM achieved the best performance (RMSE=0.0077, MAE=0.0047, R²=0.9984), outperforming both GRU and Transformer, especially under distributional shifts when prices exceeded 2400 USD. The proposed framework demonstrates the benefit of coupling scalable big data architectures with deep sequential models for financial forecasting.