Evaluating Feature Selection Techniques for Dengue Prediction with LSTM Model in Gujarat, India

Abstract

Introduction: In some specific regions of Gujarat, India, dengue fever is a remarkable public health issue. Correct forecasting of dengue is required for its control and prevention. The proposed LSTM model, this research work investigates how feature selection methods—Correlation Coefficient, Recursive Feature Elimination (RFE), and Lasso Regression—affects dengue case forecasting improvement.

Objectives: The aim of this research study is to improve the predictive performance of the LSTM model by choosing the most relevant features, reducing computational complexity, and improving accuracy.

Methods: In proposed research study three feature selection methods are used to examine key characteristics including population density, climatic factors, and historical dengue cases. LSTM model is trained on chosen features, and its performance was assessed using RMSE and R² scores.

Results: It is observed that the best performance outcome from RFE-selected features, which had an RMSE of 0.05 and a R² score of 0.85. All features have the same RMSE but a lower R² score of 0.80, it suggests the power of feature selection.

Conclusions: dengue case prediction can be improved using LSTM feature selection. By increasing model interpretability and accuracy, RFE beat other techniques and underlined the requirement for optimal input variables in time-series forecasting for public health uses.