A fuzzy model of signal detection incorporating uncertainty

A fuzzy model for signal detection in a noisy waveform was developed. Goals for the model were to incorporate uncertainty and to provide a mechanism to combine evidence that was not sensitive to estimates of probability distributions. Outcome sets were `signal present,´ `signal absent´, and `uncertain.´ The model incorporated two different parameters derived from the waveform, and a combined set membership was determined by taking the minimum of the set memberships based on the individual parameters. The model was applied to a set of brainstem auditory evoked potential waveforms, and its performance was compared to a Bayesian maximum likelihood classifier and to linear discriminant analysis. The effects of errors in estimates of the parameter statistics on classification performance were investigated. outcome set classification for the fuzzy and Bayesian models were similar to linear discriminant analysis, but the fuzzy model was less sensitive to errors in estimates of waveform parameter probabilities than the Bayesian model