This presentation deals with different data mining techniques that can be used to better detect epidemics. More precisely in this talk I first address the pattern problem extraction and show how by taking into account several dimensions (e.g. time and location), some interesting knowledge can be used to predict new cases. This part will be illustrated though real examples that have been evaluated in Neo-Caledonia. Second, I will show how additional information must be taken into account in order to improve a best prediction. More precisely I will show that remote sensing satellites images can be very helpful. I will illustrate how, by merging both supervised and supervised classification with objects extracted from satellite images it is possible to better understand the evolution of some fields and then this information coupled with patterns is particularly efficient to better predict some epidemics. Finally I will also show that other information are available in News. Usually Sentinel systems try to trigger some alarms by using the knowledge obtained by medical information (e.g. diseases, H1N1, etc). They usually perform quite well. By using information from News, they can have much more information that are not always available by the medical teams. For instance, knowing that in a country X they have many downpours can also improve the prediction. I will conclude by some visualisation tools that can be helpful for the decision maker in such a context.