Evolving fuzzy rule-based classifiers

A novel approach to on-line classification based on fuzzy rules with an open/evolving structure is introduced in this paper. This classifier can start 'from scratch', learning and adapting to the new data samples or from an initial rule-based classifier that can be updated based on the new information contained in the new samples. It is suitable for real-time applications such as classification streaming data, robotic applications, e.g. target and landmark recognition, real-time machine health monitoring and prognostics, fault detection and diagnostics etc. Each prototype is a data sample that represents the focal point of a fuzzy rule per class and is selected based on the data density by an incremental and evolving procedure. This approach is transparent, linguistically interpretable, and applicable to both fully unsupervised and partially supervised learning. It has been validated by two well known benchmark problems and by real-life data in a parallel paper. The contributions of this paper are: i) introduction of the concept of evolving (open structure) classification (eClass) of streaming data; ii) experiments with well known benchmark classification problems (Iris and wine reproduction data sets).