Evaluating the Predictive Ability of the Bipartite Dengue Contact Network Model

Abstract

This paper presents the predictive power analysis of the bipartite dengue contact (BDC) network model for identifying the source of dengue infection, defined as dengue hotspot. This BDC network model was earlier formulated, verified and validated using data collected in Sarawak, Malaysia. Then, a web-based BDC network system was implemented and subsequently tested by 7 other areas in Malaysia. The data collected using the system was then used to further evaluate the predictive ability of the BDC network model. The validity period of the dengue hotspots identified by the BDC network model was measured based on the accuracy of the predictive power analysis and Spearman’s Rank Correlation Coefficient (SRCC). Based on the results, using prior one-week data was sufficient to predict the dengue hotspot for the following week and subsequent two weeks. This shows that the hotspots are valid for two weeks. The accuracy for the outbreak areas is above 60%. Most of the model reported an SRCC above 0.70 which indicated a strong positive relationship between the hotspots in the targeted model and the validated model. Due to the accuracy and SRCC values obtained, it is suggested that the BDC network model can proceed further with retrospective data for other dengue outbreak areas in Malaysia and a prospective study for the areas that participated in this study.