Maintaining Diagnostic Knowledge-Based Systems: A Control-Theoretic Approach

Diagnostic knowledge–based systems are used in a variety of application domains to support classification decisions. The effectiveness of such systems often decreases as the application environment or user preferences change over time. Hence, frequent adjustments to the system knowledge by a human expert become necessary. We study the problem of determining the optimal amount of effort that should be exerted to maintain the system over a planning horizon (finite or infinite). Using the receiver operating characteristic curve to derive a measure for system performance, we maximize system value by balancing system benefits with maintenance costs. The problem is cast as an optimal control model in which the goal is to choose the timing and extent of maintenance that must be expended to maximize system value. We find that the optimal solution usually possesses a steady–state component. The maintenance problem is also solved as a discrete, impulse control problem, as well as one where maintenance effort has a nonlinear impact on system performance.