Parallel system and method for generating classification/regression tree

Claim 1

A classification tree generating system for generating a classification tree, in response to a training database including a plurality of properly-classified records, the tree having a plurality of nodes cach disposed in a layer relative to a root node, the tree generating system comprising: A. parallel base tree generating means including a plurality of processing nodes for generating a base classification tree having a plurality of base tree nodes each disposed in a layer relative to a base tree root node, the nodes of each layer of the base classification tree being generated in parallel based on the properly-classified records of the training database; and B. serial tree processing means including a single processor for generating a plurality of pruned classification trees in response to the base classification tree by discarding one or more nodes of the base classification tree and generating at least one selected evaluation metric for each pruned classification tree in response to an evaluation training data-base including a second plurality of properly-classified records, each evaluation metric corresponding to a number of misclassified evaluation records by the respective pruned classification tree, wherein one of the pruned trees may be selected, in response to the evaluation metrics, as the classification tree for use in classifying a plurality of as yet unclassified records of a database.

Claim 5

A method for generating a classification tree for use in classifying a plurality of records of a given database, the classification tree having a plurality of nodes, each disposed in a layer relative to a root node, the method comprising the steps of: A. generating a base classification tree defined by a plurality of base tree nodes, each disposed in a layer relative to a base tree root node, the base tree nodes being defined in parallel for each layer in response to a training database including a plurality of properly-classified records; B. labeling each base tree node, in series, with (i) a corresponding class identifier, (ii) a confidence factor that identifies the likelihood that the respective base tree node is properly associated with its corresponding class identifier, and (iii) a miscalculation cost value that corresponds to a cost associated with misclassifying records of the given data-base; C. generating a series of pruned trees from the base classification tree, each pruned tree being generated by discarding one or more base tree nodes from the base classification tree; D. calculating an evaluation value for each pruned tree in response to an evaluation database including a plurality of properly-classified evaluation records, such that one or more of the evaluation records are improperly classified by the respective pruned tree, the respective evaluation value corresponding to the number of evaluation records improperly classified by the corresponding pruned tree; E. selecting, based upon the evaluation values, one of the pruned trees as the classification tree for use in classifying the records of the given database.

Claim 6

A method for generating a classification tree for use in classifying a plurality of records of a given database, the classification tree having a plurality of nodes, each disposed in a layer relative to a root node, the method comprising the steps of: A. generating a base classification tree defined by a plurality of base tree nodes, each disposed in a layer relative to a base tree root node, the base tree nodes being defined in parallel for each layer in response to a training database including a plurality of properly-classified records; B. labeling each base tree node, in series, with (i) a corresponding class identifier, (ii) a confidence factor that identifies the likelihood that the respective base tree node is properly associated with its corresponding class identifier, and (iii) a miscalculation cost value that corresponds to a cost associated with misclassifying records of the given data-base; C. generating a series of pruned trees from the base classification tree, each pruned tree being generated by discarding one or more base tree nodes from the base classification tree; D. calculating an evaluation value for each pruned tree in response to an evaluation database including a plurality of properly-classified evaluation records, such that one or more of the evaluation records are improperly classified by the respective pruned tree, the respective evaluation value corresponding to the number of evaluation records improperly classified by the corresponding pruned tree; E. selecting, based upon the evaluation values, one of the pruned trees as the classification tree for use in classifying the records of the given database.