Name	Learning decision-oriented rules from simulation data
Theme	Machine learning for model acquisition
Begin	2002
State	Ongoing
Description	In the framework of the Sacadeau project, our aim is to build decision support systems to help catchment managers to preserve stream-water quality [2]. In collaboration with Inra researchers, three actions have been conducted in parallel [4]. * The first one consisted in building a qualitative model to simulate the pesticide transfer through the catchment from the time of its application by the farmers to the arrival at the stream. Given data on the climate over the year, on the catchment topology and on the farmer strategy, the model outputs the pesticide concentration in the stream along the year. The originality of our model is the representation of water and pesticide runoffs with tree structures where leaves and roots are respectively up-streams and down-streams of the catchment. Though Inra is the main contributor, we have participated actively to its realization. This model is now implemented and used for simulation. An in-depth analysis of many simulation results leads us to refine the model. A paper on the model has been submitted to the “Computers and Geosciences” journal. * The second action consists in identifying some of the input variables as main pollution factors and in learning rules relating these pollution factors to the stream pesticide concentration. During the learning process, we focus on actionable factors, in order to get helpful rules for decision-makers. Moreover, we take a particular interest in spatial relations between the cultivated plots and in the characteristics of crop management practices. Firstly, we decided to use Inductive Logic Programing (ILP) techniques on a simplified catchment model. The choice of ILP has been motivated by the aim to get easy-to-read and explicative rules. This first learning step (using ICL as software) showed the important impact of climate characteristics on streamwater pollution by pesticides. This work was then extended to deal with the newly implemented model and gave interesting preliminary results presented in [45]. In order to deal with the complex spatial relations existing between the catchment plots, we decided to experiment two new approaches. The first one consists in extending Inductive Learning Programming to tree structured patterns. This was done using the Aleph software which proved to be more efficient than ICL. The second approach consists in propositionalizing the learning examples and using a propositional learning process, precisely CN2. These two approaches are described and compared in [20]. * The last action consists in automatically analyzing propositional rules learned in the second step to help the experts in decision making. The aim is to go beyond the simple use of classification rules for prediction, by assisting the user in the post-analysis and in the exploitation of a large set of rules. The goal is then to find advices in order to reduce pollution whereas the learned rules are classification rules predicting if a given farmer strategy or climate leads to a polluted or not polluted situation. In 2005, we proposed the algorithm Dakar (Discovery of Actionable Knowledge And Recommendations) [87] which works as follows: starting from an unsatisfactory situation and relying on a set of classification rules, Dakar discovers a set of action recommendations and proposes them to the domain expert to improve the situation. The actions are built by selecting attributes among those describing the situation and proposing modifications on these attribute values. A heuristic evaluation function is used to evaluate the quality of the actions and rank them. We describe this work and compare it with other action recommendation systems in a paper submitted to the Journal of Machine Learning Research. This year, visualization techniques have been developed to help the experts in dealing with a large set of rules. It is now possible to locate, on a catchment map, the examples explained by the rules and thus to compare them. It is possible, for instance, to select a zone on the catchment map, and visualize the rules explaining their pollution degree.