Modeling Skewed Class Distributions by Reshaping the Concept Space

Authors: Kyle Feuz, Diane Cook

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate ICC and analyze alternative decomposition methods on well-known machine learning datasets as well as new problems in pervasive computing. Our results indicate that ICC performs as well or better than existing approaches to handling class imbalance.
Researcher Affiliation Academia Kyle D. Feuz School of Computing Weber State University Diane J. Cook School of Electrical Engineering and Computer Science Washington State University
Pseudocode No The paper describes the process steps but does not provide a formal pseudocode block or algorithm listing.
Open Source Code Yes Our source code and binary jar files are available1as Weka add-on packages. 1http://icarus.cs.weber.edu/ kfeuz/weka/
Open Datasets Yes Twelve of the datasets come from the UC-Irvine Machine learning repository (Lichman 2013)
Dataset Splits Yes We run 10 iterations of 3-fold cross validation for each dataset to determine which results are significant.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running experiments.
Software Dependencies No The paper states that its code is available as 'Weka add-on packages' but does not specify a version number for Weka or any other software dependencies.
Experiment Setup Yes We consider three different ways of selecting labels to decompose: ICC One, ICC Maj, and ICC All. Additionally, we have three different techniques for determining the number of clusters. The first technique, ICC Avg... The second technique, ICC Fix, uses a fixed number of clusters per class. This can be useful when a domain expert has knowledge about the classes and knows that each class is really composed of x sub-classes. ... We evaluate ICC using three different clustering algorithms: k-means++ (Arthur and Vassilvitskii 2007), Expectation Maximization Clustering, and Cascade Simple KMeans (Cali nski and Harabasz 1974). ... We run 10 iterations of 3-fold cross validation for each dataset to determine which results are significant.