A fusion learning breakthrough for influenza forecasting: Multivariate decomposition integrated with regulated kernel ridge intelligence
Abstract
Forecasting influenza-like illness (ILI) is critical for public health management, enabling timely actions and resource allocation to mitigate the consequences of epidemics. This study proposes RWKRidge, a novel fusion model that combines a regulated generalized ridge model (GRM) with kernel ridge regression. The model is also hybridized with the GRM via a weighted geometric mean relationship and optimized using the MS-DEPSO method. The proposed model is utilized to predict ILI rates in three areas of the United States: Mountain, South Atlantic, and West South-Central. This study presents a feature selection technique using recursive feature elimination (RFE) in conjunction with the Ridge model. Additionally, it is designed an optimum multi-variate variational mode decomposition (OMVMD) to enhance prediction accuracy. The findings indicate that the OMVMD-RWKRidge-GRM model outperforms four other machine learning (ML) models (i.e., least squares support vector machine (LSSVM), Deep extreme learning machine (DELM), long short-term memory (LSTM), and least absolute shrinkage and selection operator (LASSO)) in predicting ILI rates. It gained R values of 0.989 (training) and 0.992 (testing), along with the lowest RMSE and MAPE measurements. Based on the results of this comprehensive analysis, OMVMD-RWKRidge-GLM is the best approach for creating precise predictions with a minimal possibility of uncertainty. Forecasts made using the suggested technique can be relied upon by public health experts to effectively manage and mitigate the impact of influenza outbreaks, especially due to its higher accuracy in comparison to existing ML models.