查看文章

Association rule mining is one of the popular data mining techniques used in varied applications. The technique's biggest challenge is to reduce the time taken in finding the frequent patterns. Though there are various works available in literature on pattern mining, improving efficiency and scalability is still an important research issue. This paper presents a methodology using numerical approach to analyse large datasets and minimise the amount of data examined for generating candidate sets. The proposed algorithm uniquely encodes the data and the entire encoded database can be stored in main memory. This leads to the reduction in both space and time complexity. The frequent patterns are generated without any candidate set or tree construction. The performance of the algorithm is better with respect to both time and space compared to FP-growth and PFM algorithms.