Artificial neural network comparison of expert and novice problem-solving strategies.

Abstract The successful strategies of second-year medical students were electronically captured from computer-based simulations in immunology and infectious disease and were used to train artificial neural networks for the rapid classification of subsequent students' and experts' strategies on these problems. Such networks could categorize problem solutions of other students as successful or nonsuccessful > 85% of the time. These neural networks, however, performed poorly (as low as 13%) when classifying experienced immunologists' or internists' successful performances, suggesting an ability to distinguish between novice and expert strategies. The neural networks also identified a group of students who framed the infectious disease problems correctly, but had difficulty discriminating between differential diagnoses.