The EGC'2017 conference proposes a contest whose context is the management of green space for the city of Grenoble, focused on trees. The aim is to propose a model based on the data provided that would better predict the diseased trees, as well as the potential location of the disease. After getting some interesting results with standard models, our approach using a Cost-Sensitive One Against All model (CSOAA) allows us to obtain an accuracy of 0.86, a precision of 0.88, and a recall of 0.91 for the unilabel prediction, and a precision/recall micro of 0.82/0.74 though a precision/recall macro 0.66/0.46 for multilabel prediction. The extraction of knowledge for task 2 allowed us to highlight the interest of adding data about diseases and the concentration of pollution in the city.