Learning Parameters for a Knowledge Diagnostic Tools in Orthopedic Surgery
Résumé
We provide and illustrate a methodology for taking into account data for a knowledge diagnosis method in orthopaedical surgery, using Bayesian networks and machine learning techniques. We aim to make the conception of the student model less time-consuming and subjective. A first Bayesian network was built like an expert system, where experts (in didactic and surgery) provide both the structure and the probabilities. However, learning the probability distributions of the variables allows going from an expert network toward a more data-centric one. We compare and analyze here various learning algorithms with regard to experimental data. Then we point out some crucial issues like the lack of data.