Fault-tolerance Frequent Pattern Mining with Classifiers
Abstract
This paper proposes to mine fault-tolerant patterns with classifiers (CFT-FPM). With CFT-FPM: 1) one or more classifiers is (are) picked to be fixed at a set of specific values or specific ranges; 2) then, FT-FPM is used to mine patterns based on the corresponding fault-tolerance. Since the result is driven from classified data and proper fault-tolerance, patterns that are closer to reality are discovered. Compares to FT-FPM, CFT-FPM finds fault-tolerant patterns based on classifiers. Thus, patterns closer to reality are found since classifiers are added. In CFT-FPM, since the dataset is classified into n smaller groups that contain fewer records compared to the entire dataset, support thresholds are harder to be reached. Thus, compared to FT-FPM, CFT-FPM finds fewer patterns. In CFT-FPM, since the entire dataset is classified into n smaller groups, in each group, less length-k itemsets are generated and checked, thus, less length-(k+1) are generated and checked, and so on. Thus, compared to FT-Apriori, CFT-Apriori is much faster. CFT-Apriori is based on FT-Apriori and FT-Apriori is known to be very memory-consuming and time-consuming over large number of records, CFT-FPM is also very memory-consuming and time-consuming.
Collections
- OSU Theses [15752]