Interpreting uncertainty in fuzzy signal detection

Fuzzy logic can be used to construct signal detection systems that, unlike Bayesian classifiers, explicitly describe the degree of uncertainty in the detection decision. The usefulness of uncertain classifications depends on what can be done with them. Interpretation of uncertain classifications is illustrated by considering a fuzzy detector for QRS complexes that combines two algorithms that are sensitive to different types of noise. A simulation study incorporating different noise types demonstrates the effectiveness of fuzzy combination and shows that uncertainty characterizes situations when the algorithms conflict, that is, one is correct and the other is incorrect. Methods for dealing with uncertain classifications in QRS detection include utilizing higher level rule systems or past history to determine whether treating an uncertain classification as signal-present makes sense and skipping uncertain classifications without treating them as signal-present or signal-absent. The most appropriate method depends on the application.