Revista de Ciências da Computação, Volume II, Ano II, 2007, nº2 14 A Methodology for Filtering Association Rules Alzira Faria Instituto Superior de Engenharia do Porto aff@isep.ipp.pt Abstract Basket data analysis is an important issue in the area of Artificial Intelligence and Decision Support Systems. Association rules are a model that represents co-occurrence of items in a transaction according to some support and confidence measures. However, sometimes the number of generated association rules is too large to be analyzed. A methodology is presented to highlight the strongest rules, using a filter. Experiment results show that this filter is efficient and capable of making basket data analysis easier to implement. Keywords: association rules, filtering, artificial intelligence, decision support systems Resumo A análise de dados de cestos de compras é um assunto importante na área de Inteligência Artificial e Sistemas de Apoio à Decisão. As regras de associação são um modelo que representa co-ocorrência de itens numa transacção segundo determinados valores de suporte e confiança. No entanto, o número de regras geradas é, por vezes, suficientemente grande, dificultando a análise. Uma metodologia é apresentada para evidenciar as regras mais fortes, usando um filtro, preservando as restantes. Os resultados experimentais mostram que este filtro é eficiente e capaz de tornar a análise de dados de cestos de compras mais fácil de realizar. Palavras-chave: regras de associação, filtragem, inteligência artificial, sistemas de apoio à decisão 1-Introduction Knowledge Discovery in Databases (KDD) [Fayyad et al., 1996a] is a field that mixes the concepts of Artificial Intelligence and Decision Support Systems. Its main goal is to discover hidden useful information inside large databases, in order to transform static data into knowledge. This task is achieved by using algorithms that manipulate data and find either patterns inside data (description) or forecast unknown values (prediction), after some previous transformation of data (pre-processing). The KDD process is divided into several steps that may be seen as cyclic [Fayyad et al., 1996b]. Before applying algorithms to data, we must select the subset of data where we want to perform the discovery. Then selected data is pre-processed to remove noise and outliers. The resulting data is transformed to fit a particular input for a given algorithm. According to [Pyle, 1999], this part is essential for good results on the KDD process. The application of algorithms to the