ASSOCIATIVE CLASSIFICATION APPROACH FOR PREDICTION TARGET DATA IN THE LARGE DATA

摘要

The present invention relates to an associative classification approach for predicting target data in large-scale data, including: a step of normalizing the data; a step of generating a reference point for separation from the median of the average of target class labels and the average of the remaining class labels; a step of converting a database having a frequent item to a frequent pattern tree that has node frequency information and extracting association rules that satisfy a user threshold by measuring, based on the patterns generated, a rule complex measure that is the strong reliability rules (SRR) of the lift, leverage and conviction of each of the class labels; a step of generating reference points for multi-class label classification from the standard deviations of the target class labels and the remaining class labels from the standard deviations of the target class labels and the remaining class labels; a step of allocating predicted data input to the multi-rule set class that has the highest conformity using the average of the reference points for multi-class label classification; a step of removing extended attribute conditions from the association rules when the antecedent rules of the association rules meet the attribute conditions; and a step verifying whether it is possible to classify and predict the predicted data according to the association rules. Accordingly, the present invention is capable of preventing, by using the SRR measure, the generated rules from being biased or any wrong rule from being included.