FEPDS: A Proposal for the Extraction of Fuzzy Emerging Patterns in Data Streams

Abstract

Nowadays, most data is generated by devices that produce data continuously. These kinds of data can be categorized as data streams and valuable insights can be extracted from them. In particular, the insights extracted by emerging patterns (EPs) are interesting in a data stream context as easy, fast, and reliable decisions can be made. However, their extraction is a challenge due to the necessary response time, memory, and continuous model updates. In this article, an approach for the extraction of EPs in data streams is presented. It processes the instances by means of batches following an adaptive approach. The learning algorithm is an evolutionary fuzzy system where previous knowledge is employed in order to adapt to concept drift. A wide experimental study has been performed in order to show both the suitability of the approach in combating concept drift and the quality of the knowledge extracted. Finally, the proposal is applied to a case study related to the continuous determination of the profiles of New York City cab customers according to their fare amount, in order to show its potential.