Rodriguez et. al.'s method for constructing an ensemble of trees using random subspaces and principal components transformation applied to the input data. (weka.classifiers.meta.RotationForest). See:
Juan J. Rodriguez, Ludmila I. Kuncheva, Carlos J. Alonso (2006). Rotation Forest: A new classifier ensemble method. IEEE Transactions on Pattern Analysis and Machine Intelligence. 28(10):1619-1630. (http://doi.ieeecomputersociety.org/10.1109/TPAMI.2006.211)
Thanks to Juan Rodriguez for this contribution.
Alternating decision trees extended to handle multi-class problems. (weka.classifiers.trees.LADTree). See:
Geoffrey Holmes, Bernhard Pfahringer, Richard Kirkby, Eibe Frank and Mark Hall (2001). Multiclass alternating decision trees. Proceedings of the European Conference on Machine Learning. p 161-172. Springer.
Access to the LibLINEAR library for fast linear support vector machines and logistic regression. (weka.classifiers.functions.LibLINEAR)
Thanks to Benedikt Waldvogel for contributing this wrapper.
K-means (weka.clusterers.SimpleKMeans) now has an option to use the Manhattan distance function in combination with the component-wise median as the cluster centroids.
The sequential scatter search algorithm. (weka.attributeSelection.ScatterSearchV1). See:
Felix Garcia Lopez (2004). Solving feature subset selection problem by a Parallel Scatter Search. Elsevier.
Thanks to Adrian Pino for this contribution.
Information about a classifier and clusterer (in addition to its options) is now available from the command line by supplying the -info or -synopsis flag (in conjunction with the -h flag).
Averaged AUC, f-measure, precision, recall etc. are now available from the command line as well as in the Explorer and Experimenter GUIs.
=== Detailed Accuracy By Class === TP Rate FP Rate Precision Recall F-Measure ROC Area Class 1 0 1 1 1 1 Iris-setosa 0.96 0.04 0.923 0.96 0.941 0.992 Iris-versicolor 0.92 0.02 0.958 0.92 0.939 0.992 Iris-virginica Weighted Avg. 0.96 0.02 0.96 0.96 0.96 0.994
Usability of the KnowledgeFlow has been improved with a revamped status/log area.
The Classifier component in the KnowledgeFlow is now multi-threaded and can learn models on multiple cross-validation folds concurrently.
Support for import of PMML models (regression, general regression and neural networks) has moved into the main code base for Weka (weka.core.pmml and weka.classifiers.pmml.consumer). More information on Weka's support for PMML can be found here
Weka now logs from the main GUIs to a central file ($HOME/weka.log).