Evolutionary Online Data Mining: An Investigation in a Dynamic Environment

Recently, traditional data mining algorithms are challenged by two problems: streaming data, and changes in the hidden context. These challenges emerged from real-world applications such as network intrusion detection, credit card fraud detection, etc. Online or incremental learning becomes more important than ever for dealing with these problems. This chapter investigates XCS, an evolutionary learning classifier system, that offers an incremental learning ability and also is able to handle an infinite amount of continuously arriving data. XCS has been tested on many data mining problems and demonstrated as a potential online data mining approach. Most experiments with XCS assume a static environment. Since environments are more likely to be dynamic in real life, noise and environmental factors need to be taken into account in a good data mining approach. This chapter investigates XCS in dynamic environments, in the presence of noise in the training data. An essential requirement of an algorithm in dynamic environments is to be able to recover quickly from hidden changes, while reusing previous knowledge. Our results show that XCS is capable of recovering quickly from small changes in the underlying concepts. However, it requires significant time to re-learn a model after severe changes. We propose several strategies to force the system to learn quickly after severe changes. There are adaptive learning; re-initializing the parameters; and re-initializing the population. Experiments show improvement in the predictive performance of XCS, when compared to the traditional XCS.