lntroduction: The respiratory syncitial virus is one of the most common causes of mortality in children and older adults in the world. Objective: To predict the initial week of outbreaks and to establish the most relevant climate variables using naive Bayes classifiers and receiver operating characteristic curves (ROC). Materials and methods: The initial dates of the outbreaks in children less than five years old for the period 2005-2010 were obtained for Bogotá, Colombia. We selected the climatological variables using a correlation matrix and we constructed 1,020 models using different climatological variables and data from different weeks previous to the initial outbreak. In addition, we selected models using a six-year period (2005-2010), a four-year period (2005-2008), and a two-year period (2009-2010). We obtained the best predictive models and the most relevant climatological variables to predict the outbreak using naive Bayes classifiers and ROC curves. Results: The best models were those using a two-year period (2009-2010) and week 0, with 52% and 60% of effectiveness, respectively. Humidity was the most frequent variable in the best models (62%). Conclusions: We used naive Bayes classifiers to establish the best models to predict correctly the initial week of the outbreak. Our results suggest that the best models used humidity, wind speed and minimum temperature in outbreaks prediction.
Keywords: Bayes theorem; climatology; epidemics; forecasting; respiratory syncytial viruses.