Hybrid Machine Learning Approach to Zero-Inflated Data Improves Accuracy of Dengue Prediction.

Area of Science:

• Epidemiology
• Machine Learning
• Public Health

Background:

• Spatiotemporal dengue forecasting using machine learning (ML) aids in developing outbreak prevention strategies.
• Zero-inflated data in dengue incidence lowers prediction accuracy, posing a challenge for ML models.
• Understanding the impact of data resolution on prediction accuracy is crucial.

Purpose of the Study:

• To investigate how spatiotemporal data resolutions influence dengue incidence prediction accuracy using ML models.
• To compare the effects of spatiotemporal resolution on quantitative versus qualitative dengue predictions.
• To enhance the accuracy of dengue incidence prediction with zero-inflated data.

Main Methods:

• Compared prediction accuracy across six spatiotemporal resolutions using six ML algorithms.
• Employed data from 2009-2012 for training and 2013 for validation.
• Developed a hybrid approach for quantitative prediction, conditional on qualitative model output for zero-inflated data.

Main Results:

• Higher resolution data exhibited zero-inflation, hindering quantitative ML pattern extraction.
• Qualitative (binary) models mitigated data distribution effects.
• The hybrid approach showed high potential for predicting zero-inflated or rare phenomena like dengue.

Conclusions:

• Spatiotemporal resolution significantly impacts ML-based dengue forecasting accuracy.
• A novel hybrid qualitative-quantitative approach effectively addresses zero-inflated data challenges.
• This method offers an alternative for enhancing dengue prediction accuracy where traditional models fail.