Table 8.

Kappa statistics for training data sets for various Autoweka/Weka feature selection settings.

Therapy	No filter	cfs-best	cfs-greedy	Corr-ranker	Gain-ranker	j48-ranker	j48-greedy	CVN (lh)
Yesnoyes	0.30 (rf)	0.52 (mp)	0.52 (mp)	0.23 (lo)	0.38 (smo)	0.33 (lwl)	0.58 (mp)	0.52 (2)
Nonoyes	0.35 (sl)	0.34 (nb)	0.15 (rf)	0.22 (smo)	0.16 (lwl)	0.35 (bn)	0.15 (rf)	0.58 (2)
Nonono	0.09 (rf)	0.64 (bn)	0.5 (smo)	0.35 (ibk)	0.35 (nb)	0.35 (nb)	0.50 (rf)	0.44 (4)
Yesnono	0.7 (rf)	0.53 (sgd)	0.69 (rf)	0.39 (lo)	0.56 (rf)	0.56 (rf)	0.7 (rf)	0.53 (1)
Yesyesno	− 0.09 (dt)	0.36 (rc)	0.05 (nb)	0.22 (lwl)	0.10 (lwl)	0.0 (ab)	0.05 (nb)	0.48 (m)
Yesyesyes	− 0.07 (nbm)	− 0.07 (ibk)	− 0.01 (ibk)	0.19 (smo)	0.14 (lwl)	0.26 (rss)	− 0.07 (ibk)	0.51 (3)
Noyesno	− 0.26 (rf)	− 0.03 (mp)	− 0.03 (mp)	0.11 (mp)	− 0.26 (smo)	− 0.22 (mp)	− 0.03 (mp)	0.41 (2)
Noyesyes (*)	0.0 (rpt)	− 0.53 (rf)	− 0.53 (rf)	0.13 (rf)	0.17 (rf)	0.23 (rf)	− 0.54 (rf)	0.60 (m)

Therapy class (RAD, CHE, HOR). lh lookahead number or manually determined (m). Legend for autoweka methods: rf random forest, mp multilevel perceptron, nb Naive Bayes, bn Bayes Net, sgd stochastic gradient descent, rc random committee, ibk k-nearest neighbour classifier, sl simple logistic, nbm Naive Bayes Multinomial, rpt Fast Decision Tree REPTree (C4.5), smo fast training support vector machine, lo Logistic, lwl Locally Weighted Learning, ab AdaBoostM1, rss random subspace, dt decision table. (*) result for the validation dataset.