Predictor model for Aedes mosquito with microclimate effect using machine learning algorithms

Dengue remains a major public health threat in Malaysia, with a marked rise in cases and fatalities in 2023. While machine learning has been widely applied for dengue outbreak prediction, fewer studies have focused on predicting adult mosquito abundance, the primary vectors of dengue, particularly at the community level. This study aimed to forecast the abundance of adult female Aedes mosquitoes using microclimatic data and machine learning models, including Neural Network (NN), Support Vector Regression (SVR), Random Forest Regression (RFR), and Multiple Linear Regression (MLR). Two study sites were selected: Kolej Jasmine (a dengue hotspot in urban Bukit Raja, Petaling) and Kolej Dahlia (a non-dengue hotspot in suburban Jeram, Kuala Lumpur), both occupied by university staff and students. Environmental (temperature, relative humidity, rainfall), entomological (Adult Index [AI], Dengue Positive Trap Index [DPTI]), and epidemiological (dengue case) data were collected, with 20 gravid oviposit sticky (GOS) traps installed at each site and NS1 kits used to detect dengue virus in trapped mosquitoes. Only Aedes albopictus (Ae. albopictus) was found at the non-dengue site, while both Aedes aegypti (Ae. aegypti) and Ae. albopictus were present at the dengue hotspot. Models were developed using two approaches: (i) time lagged microclimate only predictors and (ii) combined time lagged microclimate and AI or DPTI, with the latter consistently yielding higher prediction accuracy. The best AI prediction models was SVR for Ae. albopictus at the non-dengue site (R² = 0.904, MAE = 0.125, RMSE = 0.177), MLR for total Aedes (R² = 0.944, MAE = 0.156, RMSE = 0.240) and Ae. aegypti (R² = 0.929, MAE = 0.081, RMSE = 0.108) at the dengue site, and SVR for Ae. albopictus at the dengue site (R² = 0.904, MAE = 0.125, RMSE = 0.177). For DPTI prediction, SVR performed best for total Aedes (R² = 0.553, MAE = 3.594, RMSE = 4.397) and Ae. aegypti (R² = 0.557, MAE = 2.384, RMSE = 3.074), while MLR was best for Ae. albopictus (R² = 0.326, MAE = 2.228, RMSE = 2.635). Key microclimate predictors varied by species and site, with rainfall and temperature being most influential overall. These results demonstrate the potential of machine learning to enhance mosquito surveillance and support proactive vector control by predicting oviposition activity and guiding timely interventions.