ICE-VDOP: an integrated clustering and ensemble machine learning methods for an enhanced vector-borne disease outbreak prediction using climatic variables

Dengue is prone to cause frequent disease outbreaks. An early outbreak prediction based on the climatic variables is essential to control the disease’s spread and avoid possible outbreaks. Existing non-seasonal outbreak classification uses a partition-based clustering method for outlier removal; however, a standard outlier detection with improved prediction performance is needed to detect outbreaks. We develop ICE-VDOP model, an Integrated Clustering and Ensemble machine-learning model for Vector-borne Disease Outbreak Prediction to classify outbreaks based on non-seasonal climatic characteristics for Bangkok and Bangladesh datasets. The work begins with identifying the clustering approach that generates a refined dataset ensuring the data points’ inherent structure is preserved and does not alter the association of climatic variables and Dengue Incidence. The partitioning-based clustering methods modify the direction of association between climatic variables and Dengue Incidence after outlier removal as it removes the essential data points. The comparative results shows that BIRCH (balanced iterative reducing and clustering hierarchy), a hierarchical clustering technique for outlier removal captures finer non-seasonal climate variables (temperature, rainfall and relative humidity) and gradient boosting (GB) achieves a prediction accuracy of 97% for Bangkok and 98% for the Bangladesh datasets.