Classification and sequential pattern mining from uncertain datasets.

摘要

Several research projects explore the application of uncertain databases which contain probabilistic attributes. Uncertainty in data can be caused by inherent randomness, imprecision in measuring equipment, ambiguity, information extraction from unstructured data, etc.;The classification and Sequential Pattern Mining (SPM) of uncertain datasets both play a vital role in decision making systems and have recently attracted significant attention. In this study, we propose two novel algorithms for the aforementioned problems. Our novel associative classifier for uncertain data, UAC, has an effective rule pruning strategy. Using a general model for uncertainty, our experiments show that in many cases, UAC reaches higher accuracies than the existing algorithms.;In SPM for uncertain data, other studies aimed to solve the problem for specific uncertainty models. We introduce UAprioriAll which conducts SPM from datasets with general attribute level uncertainty. Our experiments show that this method scales linearly when increasing the number of transactions.