Table 2.

Summary of the machine learning algorithms. Abbreviations are the names found in the WEKA package. Classification schemes are more general categories (types) of the algorithms. (* Func. is short for “Function”.)

Name (Abbreviation)	Class. Scheme	Details
Naïve Bayes (NaiB)	Bayes	This algorithm is based on the Bayes theorem and the assumption of the independence of all attributes. The samples are examined separately and the individual probability of belonging to a class is calculated for each particular class. Standard options were used in WEKA NäiveBayes node [14].
FilteredClassifier (Fil)	Meta	The algorithm is running an arbitrary classifier on data that has been passed through an arbitrary filter. Attribute selection filter was used with CfsSubset Evaluation and the best first search method [15].
lBk, k-nearest neighbour (lBk)	Lazy	One of the simplest algorithms, where the class membership is assigned based on the majority vote of the k-nearest neighbours of an instance. Euclidean distance was used as distance measure and k = 1 was the number of used neighbours [16].
HyperPipe (Hip)	Misc	Fast and simple algorithm, which is working well with many attributes. The basic idea of the method is the construction of pipes with different pattern of attributes to each class. The samples are monitored and selected to each class based on the pipes and the corresponding class [17].
MultiboostAB (Mboo)	Meta	This algorithm is the modified version of the AdaBoost technique with wagging. The idea of wagging is to assign random weights to the cases in each training set based on Poisson distribution. In this case Decision stump classifier was used. The number of iteration was 10 and the weight threshold was 100. The number of subcommittees was set to 3 [18].
libSVM, library SVM (SVM)	Func.*	Support vector machine can define hyperplane(s) in a higher dimensional space to separate the classes of samples distinctly. The plane should have the maximum margin between data points. Support vectors (points) can maximize the margin of the classifier. Different kernel functions and optimization parameters can be used for the classification task with SVM [19]. In this case radial basis function (RBF) was used as the kernel.
oneR, based on 1-rule, (OneR)	Rule	This algorithm ranks the attributes based on the error rate (on the training set). The basic concept is connected to 1-rules algorithms, where the samples are classified based on a single attribute [20]. Numeric values are treated as continuous ones. In this case, bucket size was 6 (standard) for the discretizing procedure of the attributes.
Bagging (Bag)	Meta	The basic concept of bagging is the creation of different models based on the bootstrapped training sets. The average (or vote) of these multiple versions are used for the prediction of class memberships for each sample [21]. In this case the number of iterations for bagging was set to 10.
Ensemble Selection (EnS)	Meta	It combines several classifier algorithms in the ensemble selection. The average prediction of the models in the ensemble is applied for the class membership determination. The selection of the models is based on an error metric (in our case RMSE). Forward selection was used for the optimization process of the ensemble. Iterations (here, 100) are also carried out such as in the case of Bagging.
Decorate (Dec)	Meta	It is also an ensemble-type algorithm, where the ensembles are constructed directly with diverse hypotheses with the application of additional artificially-constructed training examples to the original one. The classifier is working on the union of the original training and the artificial data (diversity data). The new classifiers are added to the ensemble, if the training error is not increased [22]. Several iterations are carried out to make the prediction stronger. Here, we applied 10 iterations.
Random Forest (RF)	Trees	Random forest is a tree-based method, which can be used for classification and regression problems alike. The basic idea is that it builds many trees and each of them predicts a classification. The final classification is made by a voting of the sequences of trees. The trees are weak predictors, but together they produce an ensemble; with the vote of each tree, the method can make good predictions [23].