Spectral classification using fuzzy feature sampling

Classifying biomedical spectra is often difficult due to the bir voluminous nature; typically, only a small subset of spectral features is discriminatory, while the large majority tends to have a confounding effect on pattern classifiers. We present a two-pronged approach to dealing with this issue. First, we describe an iterative technique whereby many classifier instances operate on different feature subsets. A fuzzy feature sampling method is used to identify discriminatory feature subsets. Second, subsets are aggregated using a fuzzy logic based method. Using a biomedical dataset, we empirically demonstrate that this two-pronged approach produces superior classification accuracies compared against a set of classifier benchmarks.