A Data-Driven, Flexible Machine Learning Strategy for the Classification of Biomedical Data
While biomedical data acquired from the latest spectroscopic modalities yield important information relevant to many diagnostic or prognostic procedures, they also present significant challenges for analysis, classification and interpretation. These challenges include sample sparsity, high-dimensional feature spaces, and noise/artifact signatures. Since a dataindependent ‘universal’ classifier does not exist, a classification strategy is needed, possessing five key components acting in concert: data visualization, preprocessing, feature space dimensionality reduction, reliable/robust classifier development, and classifier aggregation/fusion. These components, which should be flexible, data-driven, extensible, and computationally efficient, must provide accurate, reliable diagnosis/prognosis with the fewest maximally discriminatory, yet medically interpretable, features.
KeywordsEntropy Covariance Expense Dition Arena
Unable to display preview. Download preview PDF.