FP-Growth Algorithm: Mining Association Rules without Candidate Sets Generation

Over the years, due to modern technological advancement, unprecedented volume of data is been captured, and this has necessitated the need to mine such data to provide decision-based solution to non-trivial problems. Deploying an efficiently critical decision-based solution for handling such problems, require data mining algorithms. These evolving techniques emerged as an indispensable tools for pattern discovery in inventory data. With one notable technique being the application of Association Rule analysis, especially the Market Basket Analysis. However, mining association rules from large datasets can be daunting due to the volume of candidate sets generated by association rule algorithms like Apriori and ECLAT. Thus, candidate sets generated by these association rule based algorithms yield numerous rules, which contain both interesting and uninteresting ones. Hence, making interpretation overwhelming and decision-making challenging. On this note, this paper focused on demonstrating the efficiency of the FP-Growth algorithm in extracting relevant and interesting association rules for mining transaction itemsets over large datasets. By examining the FP-Growth algorithm design, functionality, and performance in depth analysis. The FP-Growth algorithm, which is an improved version of the Apriori algorithm is introduced with the intent to reduce the overhead costs by employing the FP-Tree data structure that efficiently encode the frequency information of itemsets in a dataset. To demonstrate performance improvement of the FP-Growth over the Apriori algorithms, the two algorithm were implemented on the WEKA data mining platform using a supermarket dataset. The performance of both algorithms is evaluated and compared in terms of computational time. The experimental results shows that the FP-Growth algorithm recorded 82.04% improvement over the Apriori algorithm. This improvement is attributed to the FP-Growth algorithm single dataset scan and the absence of candidate set generation which is inherent in the Apriori algorithm.