Coordinated multi-procedural architecture for probabilistic knowledge discovery

Current data mining procedures utilize operating cycles that encompass selection of the input data learning subset, filtering, application of the chosen mining algorithm, presentation of the output knowledge, and evaluation of the system performance on the unseen test data subset. However, a single pass through the above phases cannot achieve an appreciable result. We introduce an architecture that envelops a collection of semiautonomous multiple procedures (agents) coupled and coordinated through the common blackboard data repository. The architecture includes an efficient performance evaluation tool that assists in tuning systems parameters during multiresolutional construction of diverse elicited models. The presented compositional architecture is application specific and oriented towards probabilistic knowledge discovery.