Extracting knowledge from temporal clusters for real-time clustering

Real-time anomaly detection is an increasing need as the desire for data analysis grows along with the amount of data stored on computers. Increasing the number of features used as input and the number of clusters created also increases the time needed to classify new inputs using clustering methods. Without intelligent reduction of the search space, the need to cluster in real-time forces the user to either limit the number of input features or reduce the number of total clusters created. However, the intelligent reduction of the search space can result in real-time anomaly detection without the loss of accuracy in classification. A hybrid neural network technique that allows for clustering of sequences, the extraction of regular grammars, and a method for using the grammars for real-time classification has been developed and is presented in this paper