Deep recurrent models for forecasting infectious diseases

Infectious diseases present significant challenges to global healthcare systems due to their rapid spread and associated profound health implications. Early detection of unusual increases in case numbers is crucial for achieving efficient resource allocation and effective response planning.

Therefore, this research proposes and develops a time series predictive framework based on long short-term memory (LSTM), bidirectional LSTM (BiLSTM), and gated recurrent unit (GRU) neural network models to forecast the number of COVID-19 cases in Saudi Arabia and detect any unusual increase in cases. Google Trends and time series data for search terms, including "fever," "COVID," and "cough," serve as input, enabling models to detect the temporal patterns associated with a surge in cases. The framework is specifically designed to model temporal dependencies in sequential data, allowing the identification of early signs of anomalies in COVID-19 case trends. Therefore, we propose training the models on preprocessed time series data while adjusting for time lags to improve predictive accuracy. Evaluations of performance are conducted using mean square error (MSE) and F1-score metrics.

The experimental results demonstrate that BiLSTM returns the highest F1-score of 0.83 for the term "COVID", while LSTM and GRU reach 0.73 and 0.77, respectively. Moreover, BiLSTM outperforms LSTM and GRU at all early time lags for the search terms "fever" and "cough". The results reveal the F1-scores for the term "fever" to be 0.77, 0.62, and 0.5 for BiLSTM, GRU, and LSTM, respectively. Whereas, the F1-scores for the search term "cough" are 0.62, 0.62, and 0.5 for BiLSTM, GRU, and LSTM, respectively. Although BiLSTM incurs higher computational costs, LSTM and GRU offer efficient alternatives to deliver rapid execution. These results highlight the effectiveness of deep learning models in instances of early anomaly detection, supporting timely healthcare interventions and advancing the development of real-time monitoring systems.