Data Mining with Evolutionary Algorithms: Research Directions

Article excerpt

The Workshop entitled "Data Mining with Evolutionary Algorithms: Research Directions" was jointly sponsored by the American Association for Artificial Intelligence and GECCO-99. The general goal of the workshop was to discuss research issues concerning the integration of two areas: (1) data mining and (2) evolutionary algorithms. The workshop brought together people from these two research communities, from both academia and industry. This half-day workshop was attended by approximately 40 participants and consisted of 5 paper presentations, as follows:

First, James Thomas and Katia Sycara (both of Carnegie Mellon University) presented a genetic programming-based system for trading rule discovery, whose performance was evaluated over real-world exchange rate data in the dollar-yen and dollar-- dm markets. They focused on the issues of rule complexity and how to fight overfitting. Gary Weiss (Rutgers University) addressed the problem of predicting rare events from a sequence of events. The author has developed a genetic algorithm-based system, TIMEWEAVER, that, given a prespecified target event, learns to identify patterns in the data that successfully predict the future occurrence of this event. In essence, TIMEWEAVER uses a Michigan-style genetic algorithm to evolve a set of prediction rules. It also uses a niching strategy to ensure that a diverse set of rules is achieved.

Second, Cesar Guerra-Salcedo and Darrell Whitley (both of Colorado State University) proposed the use of genetic algorithms to select features for an ensemble of classifiers. They have experimented with two classifiers-(1) C4. …


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.