Human-Oriented Information Acquisition in Sequential Pattern Classification: Part I—Single Membership Classification

Information acquisition strategies which incorporate human heuristics are formulated for pattern classification tasks, and their effectiveness is evaluated. The heuristics are based on two observations. 1) Human decisionmakers tend to limit themselves to a subset of classes and to select features oriented toward this subset only. 2) Human decisionmakers typically employ considerations related to the history of the process, that is, to class probabilities in earlier stages (while classical Bayesian strategies consider only the current class probabilities). These heuristics are incorporated in four different strategies with which we experimented. In strategy 1 a subset is selected dynamically based on the current and earlier probabilities. In strategy 2 a subset is generated as in strategy 1, except that it is not changed dynamically but is rather retained for several steps reflecting the anchoring phenomenon with human decisionmakers. Strategy 3 (4) considers the most probable (triggered) class against the rest of the classes as a collective alternative. These strategies were compared with the classical myopic strategy 0 in which all of the classes are considered simultaneously. Extensive experiments with binary features reveal that strategies 0 and 1 are significantly superior to strategies 2–4. However, no significant difference was detected between strategy 0 and strategy 1. These findings are useful for the development of decision aids whose information selection strategies may be “tuned” to the operator's information selection behavior by offering the operator an aid which reflects his own information priorities. As a training tool, such compatibility between human and machine will enhance the effect of learning on the user who will gradually adapt himself to practicing his own style in an optimal fashion, or correct an inefficient style to a better style which is feasible within human limitations.