Several areas of knowledge produce quantitative temporal data, which demand the development techniques to identify patterns of them. Identification of emerging and decaying patterns are important in many applications because they can indicate trends that require decision-making or interventional measures. However, the literature have few researches about these kinds of patterns. This article proposes an approach for mining emerging and decaying patterns from quantitative temporal data sets using a genetic algorithm. Rules representing implications of temporal episodes are encoded into chromosomes of the genetic algorithm and these chromosomes are evolved by genetic operators. The quality of each chromosome is evaluated based on how exactly the occurrence frequency of a rule fits a straight line induced by linear regression. The decision whether the pattern is either emerging, decaying, or have no trend is taken based on the straight line slope and the regression coefficient. To prevent that the genetic algorithm converges around a single solution was used a diversity preserving method. Experiments with three quantitative temporal databases show that the results are promising, allowing to mine various emerging and decaying patterns in a single execution of the method.
展开▼
