The mean prediction accuracies of BruteDL over 25 trials on different data sets are given in Table 2. Table 3 shows the standard deviations of BruteDL's prediction accuracies over the 25 trials. The prediction accuracy is calculated as
where:
number of correct predictions on testing data;
number of testing examples.
Table 2: Mean Prediction Accuracies (%) of BruteDL on Different Data Sets
Table 3: Standard Deviation of the Prediction Accuracies of BruteDL on
Different Data Sets
As can be seen, the prediction accuracy of BruteDL gradually increases as the training set size grows in most domains. However, there are significant leaps between the prediction accuracy while using 10% as training data and that while using 20% as training data on glasses and iris data. The differences between the prediction accuracies of BruteDL over 10% and 20% training sets of cancer, lenses, soybean and monks #1 data are also quite obvious. Later we will explain that these changes are due to the characteristics of the data sets and the performance of the default rule.
