Skip to Main Content
Data mining is a constantly growing area. More and more domains of the daily life take advantage of the available tools (medicine, trade, meteorology, ...). However, such tools are confronted to a particular problem: the great number of characteristics that qualify data samples. They are more or less victims of the abundance of information. Sat domain benefits from the appearance of powerful solvers that can process huge amounts of data in short times. This paper proposes to solve supervised learning problems expressed as Sat ones. This is done to take advantage of an existing environment that allows experimenting different heuristics, such as: tabu search, genetic algorithm, ant colonies, etc., in order to extract solutions that satisfy a maximum number of clauses (Max-Sat problem). Finally, the best solutions are back-translated into rules that are applied to the data sets in order to verify that they really satisfy a maximum number of instances in the original learning problem.