A Comparison of Less Specific Versus More Specific Rules for Preterm Birth Prediction

A machine learning system was used to induce rule sets from three training data sets that included patients who had delivered preterm infants. Two series of experiments were performed: First, prediction for testing cases was done using more specific rules; Secondly, prediction was done on the basis of less specific rules. As a result, the error rate for prediction based on less specific rules was not worse than the prediction based on more specific rules. This result contradicts a perceived notion among medical practitioners that more specific rules are better for decision making.