Defuzzification

Defuzzification is the process of converting a fuzzy output — a degree of membership across multiple categories — into a crisp, actionable decision. In a fuzzy control system, a rule might conclude that a valve should be 'slightly open' with membership 0.6 and 'moderately open' with membership 0.4. Defuzzification collapses these overlapping degrees into a single numerical command: open the valve 37%. The most common methods include the centroid method, the mean of maxima, and the weighted average. But defuzzification is more than a technical step. It is the moment where a graded ontology is forced back into a discrete one. The question is not merely which method is best, but whether the need for defuzzification reveals a fundamental limitation of fuzzy systems — that even in a framework built on continuous categories, action is ultimately discrete.