Author(s) :

Arnav Bansal.

Volume/Issue :

Abstract :

Pollen allergies significantly impact global health, with rising prevalence and severity exacerbated by climate change. These conditions reduce quality of life and increase healthcare costs. Traditional pollen monitoring techniques are slow, labor-intensive, costly, and lack timely, location-specific accuracy, while general forecasts are often unreliable. This research develops a real-time prediction model for pollen allergy severity using environmental and meteorological data combined with advanced machine learning methods. Four models were evaluated: a baseline Random Forest, XGBoost, tuned Random Forest with time series cross-validation, and an advanced Random Forest incorporating cyclical date features, lagged pollen values, and rolling averages. The final model achieved the best performance with an R² value of 0.78. Significantly surpassing the approximately 50% accuracy typically achieved by prior forecasts. Results demonstrate that integrating environmental and seasonal features can substantially enhance the accuracy of pollen allergy severity predictions. Future work should aim to improve model generalizability across diverse regions and to expand the availability and temporal resolution of training data.