Mining frequent Itemsets from transaction databases using hybrid switching framework

With the growing volume of data, mining Frequent Itemsets remains of paramount importance. These have applications in various domains such as market basket analysis, clustering, classification, software bug detection web-mining to name a few. Over the recent years, several “data-structures” were employed to mine “frequent itemsets”. Unfortunately, many of them showed less efficiency in runtime or memory. This resulted in the design of Hybrid Frameworks that uses a combination of two or more data structures to extract frequent itemsets. This exploiting the benefits of different data structures while minimizing their drawbacks. This paper employs a tree-based data structure named as NegNodesets in collaboration with the list-based structure N-list for developing a novel Hybrid Framework for mining the frequent itemsets. NegNodesets have the advantage of employing bitmaps for generating a concise representation of itemsets. The N-list structure on the other hand depends on list based intersection operation for generating frequent itemsets, which is much faster than other conventional approaches. Transaction merging concept is utilized in this work to minimize the run time by merging several transactions into a single itemset. A switching criterion depends on the length of nodelist is used for switching between the algorithms. The efficacy of this approach has been enhanced by using a hash-based mechanism for generating the final set of frequent item sets. JAVA is the programming language used for coding the algorithms. The simulation analysis is carried out to know the efficacy of proposed approach in run time, memory consumption and compared with some existing approaches. From the comparative analysis, it is proved that the proposed NPLengthSwitch consumes lesser memory and run time than other techniques.