Prediction of influenza a cases in tropical climate country using deep learning model

Abstract

Influenza remains a significant public health concern particularly in tropical climate countries. Accurate prediction of influenza cases is crucial for effective resource allocation and public health planning. This study aimed to design and evaluate deep learning models for the monthly prediction of influenza A cases in a tropical climate country specifically Malaysia. The models considered both univariate and multivariate input configurations incorporating temperature and humidity variables. The study utilized a dataset spanning from January 2006 to December 2019 with training data from January 2006 to December 2016 and test data until December 2019. Various deep learning models such as RNN, LSTM, complex LSTM, GRU, Transformer and Informer were implemented and evaluated. The results demonstrated that the RNN variants specifically LSTM and GRU consistently outperformed the transformer and informer models for both univariate and multivariate prediction. The shorter input sequences (2 months) yielded better performance compared to longer sequences (12 months) with mean square error (MSE) of 0.0069, capturing more relevant temporal patterns and dependencies. The deep learning models produce a better ability to capture complex relationships and temporal patterns in the data. The findings highlight the effectiveness of the RNN variants and the impact of input configurations and sequence lengths on predictive performance. These findings provide valuable insights that can support public health planning and decision-making processes. Dataset can be obtained from the https://shorturl.at/joqxA.