Domain Independent and Extensible Pattern Recognition Through Cognitive Modeling

Abstract

Patter Recognition in time series is an indivisible part of scientific and non scientific domains. There has been a lot of research into pattern recognition in time series and research community has devised various types of techniques for the purpose. Existing pattern recognition techniques lack dynamic extensibility, as they do not provide any standard interfaces for defining new patterns for recognition dynamically. This limits the operability of these techniques for some particular domain. This research proposes a new technique for domain independent pattern recognition while giving sufficient speed and accuracy. It thus enables critical decision support systems to do pattern recognition for time series of different domains. The system emulates human visual cognition process by implementing the concept of Perceptually Important Points Identification (PIPI). Perceptually Important Points (PIP) represents minimal set of data points which are necessary to form a pattern. Another characteristic of this work is Pattern definition Language (PDL). PDL has been conceptualized for defining patterns in time series by using declarative paradigm. This enables dynamic inclusion of patterns to the system at run time. The user defines the pattern using Pattern definition markup language (PDML) which is then translated to Pattern Definition Language (PDL), before being applied on to time series for searching of patterns. PIPI prototype implementation exhibits very 10 impressive resource utilization during the detection process of the pattern and also performs highly optimized time series search.