Author(s): Il Won Seo; Soo Yeon Choi; Soo Young Park; Kyung Doug Suh
Linked Author(s): Il Won Seo
Keywords: Fecal coliform; Tree based prediction model
Abstract: Fecal coliform which cannot be measured in real-time was predicted in Daesung-ri which is located in North Han River using tree based model such as decision tree, bagging and random forest for recreational uses. Tree based model predicted fecal coliform as categorical variable which was defined according to safe fecal coliform criteria, 200 CFU/100ml. Rainfall, dam discharge and water quality explanatory variables were used to predict fecal coliform. In result, random forest model showed the best performance to predict fecal coliform and its mean total prediction error was 0.139. Further, it turned out that antecedent rainfall, antecedent rainfall duration, dam discharge, TP and EC were important variables to predict fecal coliform in Daesung-ri.