A NUMERIC APPROACH TO MINE FREQUENT PATTERNS FROM VERY LARGE DATABASE

Abstract

Mining Frequent pattern is one of the major activities in Data Mining field to extract the useful information from large databases. Today is era of big shopping complexes, mega stores, super-markets. Shopping using e-commerce portals like flipkart, amazon, snapdeal etc is increasing day by day. As a result databases of all these mega stores, shopping complexes and e-commerce portals are increasing many folds every year. Frequent patterns are used by the big corporate houses to know the interest and purchasing trend of their customers and accordingly they plan their sales or marketing strategies. This lead to need for faster algorithm to mine frequent patterns from these large databases. This research work presents a novel idea for mining frequent patterns from very large database. Proposed new algorithms i.e. Magic Number Algorithm is based on the converting the database items to Numeric equivalents. Once we have all items represented by numerals then we can apply mathematics and logic on them in easier and faster way as compared to operations on strings. Using same experimental setup and environment conditions, proposed new algorithm proves superior to Apriori Algorithm by approximate 70% performance improvement. At Minimum Support level = 4 and transaction count= 1000,000 Time taken by Magic Number Algorithm - 105 milli seconds Time taken by Apriori Algoritm - 345 milli seconds