Visitors   Views   Downloads
NOT PEER-REVIEWED
"PeerJ Preprints" is a venue for early communication or feedback before peer review. Data may be preliminary.

Supplemental Information

A General Framework for Class-Specific Feature Selection

DOI: 10.7287/peerj.preprints.27740v1/supp-2

Cumulative sum of the rows of matrix C in descending order of ||•||1. The regularization parameter in eq. 7 is set to 20 and 50, respectively

DOI: 10.7287/peerj.preprints.27740v1/supp-3

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. KNN classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-4

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. KNN classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-5

Comparison of several TFS accuracies against SMBA and SMBA-CSFS on eight datasets, when a varying number of features is selected. Logistic Regression classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-6

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. Logistic Regression classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-7

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. Naive Bayes classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-8

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. Naive Nayes classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-9

Comparison of several TFS accuracies against SMBA and SMBA-CSFS on eight datasets, when a varying number of features is selected. Decision Tree classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-10

Comparison of several CSFS accuracies against SMBA-CSFS on eight datasets, when a varying number of features is selected. Decision Tree classifier with 5-fold CV was used

DOI: 10.7287/peerj.preprints.27740v1/supp-11

Supplemental Tables - KNN

In the Excel file we have reported the accuracy results on top 20 and 80 selected features by using the KNN classifier, comparing TFS vs SMBA-CSFS and GF-CSFS vs SMBA-CSFS. We also have reported the Friedman test to rank the algorithms’ performance across multiple datasets, taking into account the accuracy results achieved on subsets of 20 and 80. This involves ranking each row together and then considering the values of ranks by columns where the best performing algorithm gets the rank of 1, the second best rank 2 and so on. In addition, we have reported the Cumulative Rank(CR) considering either all the dataset or a subset of them (CR≥5). Finally, a Nemenyi post-hoc test (all vs all) is ran to compare all the other methods against SMBA-CSFS with the aim to determine which groups are significantly different, with respect to SMBA-CSFS. The significance level α is set to 5%. We have provided the ranking of the classification procedures based on TFS against SMBA-CSFS, for the top 20 and 80 features and the p-values corresponding. We also provided the ranking of the classification procedures based on GF-CSFS against SMBA-CSFS, for the top 20 and 80 features, respectively along with the corresponding p-values tables.

DOI: 10.7287/peerj.preprints.27740v1/supp-12

Supplemental_Tables - Logistic Regression

In the Excel file we have reported the accuracy results on top 20 and 80 selected features by using the Logistic Regression classifier, comparing TFS vs SMBA-CSFS and GF-CSFS vs SMBA-CSFS. We also have reported the Friedman test to rank the algorithms’ performance across multiple datasets, taking into account the accuracy results achieved on subsets of 20 and 80. This involves ranking each row together and then considering the values of ranks by columns where the best performing algorithm gets the rank of 1, the second best rank 2 and so on. In addition, we have reported the Cumulative Rank(CR) considering either all the dataset or a subset of them (CR≥5). Finally, a Nemenyi post-hoc test (all vs all) is ran to compare all the other methods against SMBA-CSFS with the aim to determine which groups are significantly different, with respect to SMBA-CSFS. The significance level α is set to 5%. We have provided the ranking of the classification procedures based on TFS against SMBA-CSFS, for the top 20 and 80 features and the p-values corresponding. We also provided the ranking of the classification procedures based on GF-CSFS against SMBA-CSFS, for the top 20 and 80 features, respectively along with the corresponding p-values tables.

DOI: 10.7287/peerj.preprints.27740v1/supp-13

Supplemental_Tables - Naive Bayes

In the Excel file we have reported the accuracy results on top 20 and 80 selected features by using the Naive Bayes classifier, comparing TFS vs SMBA-CSFS and GF-CSFS vs SMBA-CSFS. We also have reported the Friedman test to rank the algorithms’ performance across multiple datasets, taking into account the accuracy results achieved on subsets of 20 and 80. This involves ranking each row together and then considering the values of ranks by columns where the best performing algorithm gets the rank of 1, the second best rank 2 and so on. In addition, we have reported the Cumulative Rank(CR) considering either all the dataset or a subset of them (CR≥5). Finally, a Nemenyi post-hoc test (all vs all) is ran to compare all the other methods against SMBA-CSFS with the aim to determine which groups are significantly different, with respect to SMBA-CSFS. The significance level α is set to 5%. We have provided the ranking of the classification procedures based on TFS against SMBA-CSFS, for the top 20 and 80 features and the p-values corresponding. We also provided the ranking of the classification procedures based on GF-CSFS against SMBA-CSFS, for the top 20 and 80 features, respectively along with the corresponding p-values tables.

DOI: 10.7287/peerj.preprints.27740v1/supp-14

Supplemental_Tables - Decision Tree

In the Excel file we have reported the accuracy results on top 20 and 80 selected features by using the Decision Tree classifier, comparing TFS vs SMBA-CSFS and GF-CSFS vs SMBA-CSFS. We also have reported the Friedman test to rank the algorithms’ performance across multiple datasets, taking into account the accuracy results achieved on subsets of 20 and 80. This involves ranking each row together and then considering the values of ranks by columns where the best performing algorithm gets the rank of 1, the second best rank 2 and so on. In addition, we have reported the Cumulative Rank(CR) considering either all the dataset or a subset of them (CR≥5). Finally, a Nemenyi post-hoc test (all vs all) is ran to compare all the other methods against SMBA-CSFS with the aim to determine which groups are significantly different, with respect to SMBA-CSFS. The significance level α is set to 5%. We have provided the ranking of the classification procedures based on TFS against SMBA-CSFS, for the top 20 and 80 features and the p-values corresponding. We also provided the ranking of the classification procedures based on GF-CSFS against SMBA-CSFS, for the top 20 and 80 features, respectively along with the corresponding p-values tables.

DOI: 10.7287/peerj.preprints.27740v1/supp-15

Additional Information

Competing Interests

The authors declare that they have no competing interests.

Author Contributions

Davide Nardone conceived and designed the experiments, performed the experiments, analyzed the data, prepared figures and/or tables, performed the computation work, authored or reviewed drafts of the paper, approved the final draft.

Angelo Ciaramella conceived and designed the experiments, prepared figures and/or tables, authored or reviewed drafts of the paper, approved the final draft.

Antonino Staiano conceived and designed the experiments, prepared figures and/or tables, authored or reviewed drafts of the paper, approved the final draft.

Data Deposition

The following information was supplied regarding data availability:

The data supporting the experiments in this article are available at https://zenodo.org/record/2709491#.XNXkhpMzagQ.

A Python software package is available through GitHub at https://github.com/DavideNardone/A-Sparse-Coding-Based-Approach-for-Class-Specific-Feature-Selection, containing all the source codes used to run the software.

Funding

This work was supported by Dipartimento di Scienze e Tecnologie Univesrite.gli Studi di Napoli Parthenope (Sostegno alla ricercaindividuale per il triennio 2016-2018project). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.


Add your feedback

Before adding feedback, consider if it can be asked as a question instead, and if so then use the Question tab. Pointing out typos is fine, but authors are encouraged to accept only substantially helpful feedback.

Some Markdown syntax is allowed: _italic_ **bold** ^superscript^ ~subscript~ %%blockquote%% [link text](link URL)
 
By posting this you agree to PeerJ's commenting policies