AN IMPROVED ANT COLONY ALGORITHM FOR EFFECTIVE MINING OF FREQUENT ITEMS

Data Mining involves discovery of required potentially qualified content from a heavy collection of heterogeneous data sources. Two decades passed, still it remains the interested area for researchers. It has become a flexible platform for mining engineers to analyse and visualize the hidden relationships among the data sources. Association rules have a strong place in representing those relationships by framing suitable rules. It has two powerful parameters namely support and confidence which helps to carry out framing of such rules. Frequent itemset mining is also termed to be frequent pattern mining. When the combination of items increases rapidly, we term it to be a pattern. The ultimate goal is to design rules over such frequent patterns in an effective manner i.e in terms of time complexity and space complexity. The count of evolutionary algorithms to achieve this goal is increasing day by day. Bio Inspired algorithms holds a strong place in machine learning, mining, evolutionary computing and so on. Ant Colony Algorithm is one such algorithm which is designed based on behaviour of biological inspired ants. This algorithm is adopted for its characteristic of parallel search and dynamic memory allocation. It works comparatively faster than basic Apriori algorithm, AIS, FP Growth algorithm. The two major parameters of this algorithm are pheromone updating rule and transition probability. The basic ant colony algorithm is improved by modifying the pheromone updating rule in such way to reduce multiple scan over data storage and reduced count of candidate sets. The proposed approach was tested using MATLAB along with WEKA toolkit. The experimental results prove that the stigmeric communication of improved ant colony algorithm helps in mining the frequent items faster and effectively than the above stated existing algorithms.