MIT Press

Table 8

Results from our 15 meta-classifiers applied to Toefl11, using both discrete and continuous outputs from the base classifiers. The best result in each column is in bold. The results can be compared with the baselines in Table 7.

Meta-classifier	Discrete		Continuous
Meta-classifier	CV	Test	CV	Test
Random Baseline	9.1	9.1	9.1	9.1

Linear SVM	84.3	84.4	84.5	85.2
RBF-Kernel SVM	84.2	84.7	84.6	85.1
Logistic Regression	83.9	83.8	84.3	84.8
Ridge Regression	84.3	84.8	84.2	84.5
Perceptron	78.5	81.5	80.9	81.7

Decision Tree	77.6	78.3	75.1	75.2
QDA	56.8	57.3	67.4	67.9
LDA	84.3	84.7	85.2	86.8

Nearest Centroid	83.6	83.5	83.1	83.5
5-NN	83.2	82.5	81.6	82.0
10-NN	83.5	83.0	83.0	84.5
15-NN	83.6	83.4	83.2	84.3
20-NN	83.6	83.6	83.3	84.7
50-NN	83.4	83.6	83.7	84.2
100-NN	83.1	83.3	83.6	84.1

Meta-classifier	Discrete		Continuous
Meta-classifier	CV	Test	CV	Test
Random Baseline	9.1	9.1	9.1	9.1

Linear SVM	84.3	84.4	84.5	85.2
RBF-Kernel SVM	84.2	84.7	84.6	85.1
Logistic Regression	83.9	83.8	84.3	84.8
Ridge Regression	84.3	84.8	84.2	84.5
Perceptron	78.5	81.5	80.9	81.7

Decision Tree	77.6	78.3	75.1	75.2
QDA	56.8	57.3	67.4	67.9
LDA	84.3	84.7	85.2	86.8

Nearest Centroid	83.6	83.5	83.1	83.5
5-NN	83.2	82.5	81.6	82.0
10-NN	83.5	83.0	83.0	84.5
15-NN	83.6	83.4	83.2	84.3
20-NN	83.6	83.6	83.3	84.7
50-NN	83.4	83.6	83.7	84.2
100-NN	83.1	83.3	83.6	84.1

View Large

This Feature Is Available To Subscribers Only