<p>Accurate forecasting of stock prices, pollution levels, and energy consumption is crucial for ensuring economic stability, public health, and efficient resource management. This paper introduces a novel hybrid model that combines Dual Seasonal Exponential Smoothing (DSES) with a Hierarchical Attention-based Encoder-Decoder Bidirectional Gated Recurrent Unit (HA-ED-BiGRU), further optimized by the Nipuna Activation Function (NAF). The model effectively captures seasonal patterns and trends in stock, pollution, and energy data while leveraging Deep Learning (DL) to handle complex temporal dependencies. The NAF reduces gradient saturation, contributing to more stable and robust predictions. The proposed DSES-HA-ED-BiGRU-Nipuna model is evaluated against state-of-the-art models like LSTM, BiLSTM, and ED-BiGRU using datasets from stock prices, pollution levels, and energy consumption. Performance is measured through metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Coefficient of Determination (R²), Accuracy, Directional Accuracy (DA), Win/Loss Ratio (WLR), and Theil’s U-Statistic (TUS). Experimental results demonstrate that the proposed model outperforms existing methods, delivering superior accuracy and predictive reliability. The integration of DSES with the HA mechanism improves the model’s ability to handle multi-seasonal patterns and enhances interpretability, proving a robust framework for time series forecasting in various domains, including stock prices, pollution, and energy consumption.</p>