Academic
Publications
Ensemble Feature Selection Based on Contextual Merit and Correlation Heuristics

Ensemble Feature Selection Based on Contextual Merit and Correlation Heuristics,10.1007/3-540-44803-9_13,Seppo Puuronen,Iryna Skrypnyk,Alexey Tsymbal

Ensemble Feature Selection Based on Contextual Merit and Correlation Heuristics   (Citations: 4)
BibTex | RIS | RefWorks Download
Recent research has proven the benefits of using ensembles of classifiers for classification problems. Ensembles of diverse and accurate base classifiers are constructed by machine learning methods manipulating the training sets. One way to manipulate the training set is to use feature selection heuristics generating the base classifiers. In this paper we examine two of them: correlation-based and contextual merit -based heuristics. Both rely on quite similar assumptions concerning heterogeneous classification problems. Experiments are considered on several data sets from UCI Repository. We construct fixed number of base classifiers over selected feature subsets and refine the ensemble it-eratively promoting diversity of the base classifiers and relying on global accuracy growth. According to the experimental results, contextual merit-based ensemble outperforms correlation-based ensemble as well as C4.5. Correlation-based ensemble produces more diverse and simple base classifiers, and the iterations promoting diversity have not so evident effect as for contextual merit-based ensemble.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
    • ...Other ”favorite class” feature selection iterative methods have been considered by Puuronen at el. [13] - they also derived special contextual merit measures instead of using the simple correlation...

    Jerzy Stefanowski. Combining Answers of Sub-classifiers in the Bagging-Feature Ensembles

    • ...In this chapter we consider correlation -based feature selection approaches described in the literature and then present how to construct an ensemble of classifiers using them with the help of the EFS-corr (Ensemble Feature Selection guided by correlation-based heuristics) algorithm developed in [7]...
    • ...The algorithm EFS-corr is a modification of the algorithm EFS-ref considered earlier in [7]...
    • ...In general, each data set has its own optimal threshold value for the feature merits [7,8]...

    Iryna Skrypnyklet al. Selection of Voice Features to Diagnose Hearing Impairments of Childre...

    • ...Comparisons between CM and CR have been presented in our earlier paper [11] where we reported that CM initial ensemble results in higher average accuracy with 6 data sets (2 statistically significant)...
    • ...In [11] we presented similar comparison between the initial and final ensembles of CM and CR. In both cases the final ensemble resulted in higher accuracy for 7 data sets but the difference over all the data set was not statistically significant...

    Seppo Puuronenet al. Correlation-Based and Contextual Merit-Based Ensemble Feature Selectio...

Sort by: