A medical database with 257 patients thought to have acute appendicitis has been analyzed.Binary classifiers composed of very simple univariate if-then classification rules (1R rules) were synthesized,and are shown to perform well for determining the true disease status.Discriminatory performance was measured by tyhe area under the receiver operating characteristic (ROC) curve.Although an 1R classifier seemingly performs slightly better than a team of experienced physicians when only readily available clinical variables are employed,an analysis of cross-validated simulations shows that this perceived improvement is not statistically significant (p<0.163).However,further addition of biochemical test results to the model yields an 1R classifier that is significantly better than both the physicians (p<0.03) and an 1R classifier based on clinical variables only (p<0.0003).
展开▼
