SW-LSTM: Strong-Weak Long Short-Term Memory Fusion Network for Forecasting Pandemic Spread
Abhishek Sharma, Saumya, Meghanshu Verma, Himanshu Singh, Badri Narayan Subudhi, Vinit Jakhetiya, Udit Satija · 2025
The pandemic created unparalleled global health challenges, resulting in millions of deaths worldwide, effectively halting the world. Various statistical and machine learning models have been previously deployed to forecast the spread of the pandemic worldwide. Forecasting the spread rate supports pandemic control by guiding the creation of strategies to limit its transmission. Despite significant efforts made to forecast the spread of such infectious diseases in India, efficient forecasting using deep learning systems is still lacking. To overcome this problem, we have proposed a novel, effective, intelligent framework for predicting the course of any infectious diseases using a novel LSTM-based fusion model. The proposed model combines two different LSTM networks, the Strong and Weak LSTM models, to improve time series forecasting accuracy. The model assigns 25 % weight to the strong model and 75% weight to the weak model, leveraging their complementary strengths for better predictions. We have used the COVID-19 India dataset. While our tests focused on COVID-19 India data, the methodology can be readily adapted to other pandemic diseases. We have proposed to accurately estimate the rise of the pandemic cases in the near future by considering various factors like; age, gender, geographical location, etc. A brief comparative study with different time-series deep-learning models is also given. The model's performance is evaluated using mean absolute percentage error (MAPE) on the COVID-19 India dataset and provides comparative results with six state-of-the-art techniques.