Modeling Factors Associated With Diarrhea Caused by Cryptosporidium Species Using Machine Learning Methods.

Abstract

OBJECTIVE: Cryptosporidium spp. is an important pathogen responsible for severe diarrheal illness, especially in children, and is transmitted by various modes. The present work is aimed at categorizing Cryptosporidium spp. infection and determining associated risk factors by ML on a known dataset of diarrhea among children. MATERIALS AND METHODS: For classification, we used random forest and bagging CART trees. Model discrimination was measured by accuracy, balanced accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and F1-score. Then, a 5-fold cross-validation method was used to verify the reliability of the model. Importance values were also calculated to identify the most important risk factors for infection. RESULTS: The bagged CART model emerged as the best among the models applied, with slightly better classification. For this model, performance metrics were: accuracy (87.2%), balanced accuracy (56.3%), sensitivity (97.2%), specificity (15.4%), positive predictive value (89.3%), negative predictive value (42.9%), F1-score (93.0%). As shown by the variable importance analysis, the strongest risk factor was the number of people in the household (people ≥ 5), which represented a higher risk of infection in crowded housings. Sources of water also came up as an important environmental factor; plain tap water and pipe-line water appeared to be major causes of transmission. CONCLUSION: Such results indicate that waterborne transmission is the main route of Cryptosporidium spp. INFECTION: These findings underscore the importance of water quality improvements, including efforts to address water disinfection, particularly in areas with household crowding and inadequate sanitation access.