Stock Price Forecasting Using a Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Recurrent Neural Networks

Stock prices are inherently volatile and influenced by multiple factors, necessitating accurate forecasting to minimize investment risks. This study proposes a hybrid architecture that integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for signal decomposition and Recurrent Neural Networks (RNN) models, namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), to effectively forecast the stock prices of PT Alamtri Resources Indonesia Tbk (ADRO). A comparative analysis is conducted to determine the most effective hybrid model for supporting investment decision making. CEEMDAN decomposition yields seven Intrinsic Mode Function (IMF) components and one residual, which are predicted using LSTM and GRU with various hyperparameter combinations and then reconstructed into an overall forecast. Evaluation using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) shows that CEEMDANLSTM outperforms CEEMDAN-GRU, achieving an MAE of 66.7775, an RMSE of 128.9119, and a MAPE of 2.3101%. The twenty days ahead forecast indicates an initial decline followed by a gradual increase with relatively stable fluctuations, confirming the effectiveness of the proposed hybrid approach in stock price forecasting.