Vol. 1, Issue 1, Part A (2024)

Accurate time-series forecasting is vital for decision-making in fields such as finance, energy, retail, and climate science, where anticipating future trends directly influences strategic planning and operational efficiency. This study presents a comprehensive comparative analysis of predictive modeling techniques, encompassing classical statistical models, machine learning algorithms, and deep learning architectures, to evaluate their effectiveness in diverse forecasting scenarios. Using benchmark datasets from multiple domains, models such as ARIMA, ETS, Prophet, Random Forest, Support Vector Regression (SVR), Long Short-Term Memory (LSTM), DeepAR, and Temporal Fusion Transformer (TFT) were assessed based on forecasting accuracy, computational efficiency, and robustness. Evaluation metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Scaled Error (MASE) were applied to ensure a fair and comprehensive performance comparison. Results indicated that while traditional models remain reliable for stationary and well-behaved data, deep learning architectures and ensemble approaches significantly outperform them when handling nonlinear dependencies, irregular seasonality, and long-term temporal correlations. The ensemble model, integrating outputs from statistical and neural approaches, demonstrated the lowest overall forecasting error and the most consistent performance across datasets. Findings support the hypothesis that hybrid frameworks leveraging the interpretability of classical methods with the adaptability of deep learning can optimize accuracy and generalization in practical applications. The study also provides practical recommendations emphasizing model selection based on data complexity, resource availability, and operational constraints. Overall, the research highlights that ensemble-based hybrid intelligence systems represent the most promising direction for scalable, accurate, and interpretable time-series forecasting in modern data-driven environments.