| The goal of this thesis is to develop an efficient association-rule mining algorithm that is suitable for large data sets with long patterns. The FP-growth algorithm is a recent Association Rule Mining (ARM) technique which efficiently extracts knowledge, such as associative patterns between attribute values of large data sets, due to its highly compact data representation and pattern finding scheme.; The proposed algorithm, the Partitioned FP-growth (PFP-growth) algorithm, involves the use of parallel processing techniques to the FP-Growth algorithm to reduce the processing bottleneck that arises when extremely large data sets are mined sequentially. The test data for the proposed algorithm is extracted from medical images (mammograms) which is a typical example of such large data sets.; Experiments show that the PFP-growth algorithm improves on the mining efficiency of the FP-growth algorithm in segments prone to processing bottlenecks by achieving between 23.20% to 45.07% speed up, indicating a positive contribution with the use of parallel techniques. Also, processing speeds show that the PFP-growth algorithm scales well with the number of records mined. The results are a set of association rules that provide a framework for an image classifier. Classifying new images with the image classifier indicates a detection accuracy of approximately 80.36%. |
