A fusion-based hybrid modeling framework for pandemic time series forecasting

Reliable epidemic forecasting is essential for effective public health planning and resource allocation in Malaysia, particularly in the context of population growth and climate variability. Traditional statistical models such as Auto Regressive Integrated Moving Average (ARIMA) capture linear patterns, while deep learning models, including Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM), effectively learn nonlinear dependencies. However, each approach in isolation struggles to adapt to abrupt changes in epidemic dynamics. This study aims to develop and evaluate a hybrid (Prophet-GRU) forecasting framework that integrates statistical and deep learning techniques to enhance predictive accuracy and robustness in modeling COVID-19 dynamics in Malaysia. Daily records of confirmed, recovered, and deceased COVID-19 cases from January to September 2021 were obtained from the Malaysian Ministry of Health. A hybrid Prophet–GRU model was proposed and evaluated against leading statistical and machine learning baselines using a chronological training–validation–testing strategy. Forecasting performance was assessed using standard error and goodness-of-fit metrics. The hybrid model demonstrated consistent predominance over single-model and others hybrid model approaches. By combining Prophet and GRU, the framework delivered remarkable forecasting performance across the majority of case categories, successfully identifying both extended trends and brief fluctuations. When measured against ARIMA and independent neural networks, the hybrid model notably diminished prediction errors, especially in active and recovered case projections. Combining the interpretability of statistical models with the adaptability of deep learning significantly improves time-series forecasting under complex epidemic conditions. The Prophet–GRU framework has showed strong potential as a decision-support tool for Malaysian public health authorities, enabling timely resource allocation and enhanced preparedness for future outbreaks.