Distributed High-Dimensional Mining Tool for Bioscience Data Analysis

We propose an innovative data mining tool to systematically analyze huge amounts of experiment and sensor data in bioscience. Our proposed tool, named ABMiner (where AB for Air-Borne) is the synergy of attribute-oriented induction, a classification ensemble, and distributed computation. Our proposed tool has three unique contributions. First, we use attribute-oriented induction for pre-clustering to overcome the difficulty of curse of dimensionality when mining a huge amount of high-dimensional data. Second, the classification ensemble combines a set of independent classifiers in some reasonable manner such that the accuracy of the ensemble is better than any single classification algorithm. In this proposal, we extend and enhance the base classifiers of the ensemble in two dimensions, namely types of algorithms and variations of data sets, to enrich the base classifiers and guarantee the accuracy of the classification model. Finally, we derive the architecture of distributed computation from the idea of two dimensions of classification ensemble for improving computational efficiency through parallel computation. In the architecture, different classification algorithms can run in different machines (nodes), and the same classification algorithm with different training data sets and guidance parameters can also run in different nodes.