Application of Least Absolute Shrinkage Selector Operator (LASSO) to Predict Dengue

Abstract

Dengue which was first detected mainly in South East Asia during 1940s is now a serious public
health concern across the subtropical and temperate regions of America, Europe and China due to the change
in global climate and international travel. This research aims to forecast dengue cases four weeks in advance
for San Juan, Puerto Rico and Iquitos, Peru using Least Absolute Shrinkage Selector Operator (LASSO)
model. Two models were built for each city to evaluate the value addition from the feature engineering process
of climatic variables. Time series model diagnostic plots were used to guide model selection between the two
sets of models. Significant climatic variables that could impact the dengue cases in each city were identified
accordingly. LASSO’s flexibility in incorporating a variety of predictors and its ease of interpretation present
LASSO as a compelling case against the general predictive models. In this paper, model M1 produced the best
prediction with Mean average error at 1.1104 for the Test set as it was better able to mimic the peaks in
comparison with the feature engineered model M2. Public health regulators could make use of such models
to facilitate the timing of vector control and public health campaigns along with the allocation of medical
resource to cope with potential dengue outbreaks.