Skip to content

Releases: SCI2SUGR/KEEL

Keel 3.0

24 Sep 11:43
@SCI2S SCI2S
3.0
8cae49b
Compare
Choose a tag to compare
Keel 3.0 Latest
Latest 
            KEEL SOFTWARE SUITE 3.0 - Open Source - V2015-03-23

KEEL 3.0 Version is available now.

New features (from v2015-03-23):

  1. A new Semi-Supervised Learning (SSL) module have been included. Significant additions related to this novel module are the following:
    1.1 Eleven datasets for semi-supervised learning created from several original UCI problems.
    1.2 Three different paradigms of classifiers with different algorithms:
    1.2.1 Multiple-classifiers: ADE_CoForest, CLCC, Co-Bagging(CoBC), CoForest, Co-Training, DE-TriTraining, Democratic-Co, Random subspace method for co-training, (RASCO), Rel-RASCO, Tri-Training.
    1.2.2 Single-classifiers: APSSC, Self-Training, SETRED, Self-training nearest neighbor rule using cut edges (SNNRCE)
    1.2.3 Supervised: C4.5, Naïve Bayes, Neural Networks and SVM with SSL
  2. For the sake of developing a better validation of the experimental results, a recent data partitioning scheme, Distribution optimally balanced stratified cross-validation (DOB-SCV), has been added. This way, the dataset shift between training and test folds is reduced and the generalization capability of the algorithms can be better identified. The standard Stratified Cross and 5x2 Validation methods are maintained.
  3. The unsupervised module of Association Rules have been extended with several algorithms from the state-of-the-art related to genetic learning: The Alatas et al. method, Alcala et al. method, ARMMGA, EARMGA, GAR, GENAR, Genetic Fuzzy Apriori, Genetic Fuzzy Apriori DC, MODENAR, and MOEA_Ghosh. Additionally, new published approaches have been also included: MOPNAR and QAR_CIP_NSGAII.
  4. The Evolutionary Fuzzy Rule Learning section have been updated with the Genetic Programming-based learning of COmpact and ACcurate fuzzy rule-based classification systems for High-dimensional problems (GP-COACH) and Interval-Valued fuzzy reasoning method with TUning and Rule Selection (IVTURS) algorithms.
  5. A complete new Fuzzy KNN section has been added with 19 new approaches, including Condensed Fuzzy K-Nearest Neighbors classifier (CFKNN), Dempster-Shafer theory based K-Nearest Neighbors classifier (D_SKNN), Fuzzy C-Means K-Nearest Neighbors classifier (FCMKNN), Fuzzy Edited Nearest Neighbor classifier (FENN), Fuzzy Rough K-Nearest Neighbors Approach (FRKNNA), Fuzzy-Rough Nearest Neighbor algorithm (FRNN), Fuzzy-Rough Nearest Neighbor classifier - Fuzzy Rough Sets (FRNN_FRS), Fuzzy-Rough Nearest Neighbor classifier - Vaguely Quantified Rough Sets (FRNN_VQRS), Fuzzy K-Nearest Neighbors classifier (FuzzyKNN), Fuzzy Nearest Prototype classifier (FuzzyNPC), Genetic Algorithm for Fuzzy K-Nearest Neighbors classifier (GAFuzzyKNN), Intuitionistic Fuzzy K-Nearest Neighbors classifier (IF_KNN), Intuitionistic Fuzzy Sets K-Nearest Neighbors classifier (IFSKNN), Intuitionistic Fuzzy Version of K-Nearest Neighbors classifier (IFV_NP), Interval Type-2 Fuzzy K-Nearest Neighbors classifier (IT2FKNN), Jozwik Fuzzy K-Nearest Neighbor algorithm (JFKNN), Pruned Fuzzy K-Nearest Neighbors classifier (PFKNN), Possibilistic Instance Based Learning (PosIBL), and Variance Weighted Fuzzy K-Nearest Neighbors classifier (VWFuzzyKNN).
  6. A new Dual Feature and Prototype Selection/Weighting algorithm has been added to "Evolutionary Prototype Selection" section. This method is "Instance and Feature Selection based on Cooperative
    Co-evolution" (IFS_COCO).
  7. The "Nested Generalized Learning" section has been updated with a recent algorithm based on Hyperrectangles, i.e. the "Evolutionary Hyperrectangle Selection based on CHC" (EHS_CHC).
  8. Three novel algorithms have been added into the "Imbalanced Learning" Module. Two of these methods are included into the "ensembles for class imbalance", i.e. AdaBoost.NC and EUSBoost. The remaining method, Hierarchical genetic programming fuzzy rule based classification system with rule selection and tuning (GP-COACH-H), is an algorithmic modification of the GP-COACH approach especially designed for class imbalance.
  9. There are two significant improvements within the "Decision Trees" section. First, both One-vs-One (OVO) and One-vs-All (OVA) binarization approaches have been added using C4.5 as baseline
    classifier. Additionally, when OVO is selected, several aggregations for the score matrix can be applied, from weighted voting, to non-dominance criterion. Second, an ensemble model
    AdaBoost.NC has been added, also using C4.5 as baseline classifier.
  10. Several corrections have been maded within the AntMinerPlus and AdvancedAntMinerPlus algorithms.
  11. Help files for the Feature Selection methods have been modified and updated.
  12. RBNF methods now provide the resulting model as output.
  13. The default parameters of SVM algorithm have been updated.
  14. New versions of the SLAVE algorithm (SLAVE, SLAVE2 and NSLV) have been included within the Evolutionary Fuzzy Rule Learning category.
Assets 2
Loading