Luminita Mihaela Moruz, Liviu Ciortuz

The field of feature selection in data mining has experienced a great development in the past few years. Unfortunately, there are few worksthat show a comparative analysis of feature selectionalgorithms introduced until now.

In this technical report we analyze the effect that feature selection has on the performance of five learning algorithms. We use nine of the most known feature selection algorithms. We conduct our study on two classification problems, Dermatology and Zoo, whose datasets are provided in the UCI (University of California atIrvine) repository.

We realize a comparative study on the effects that the chosen feature selection algorithms have on the performance and the running time of the learning algorithms used in solving the two problems.

On these problems, the feature selection algorithms provide an accuracywhich is 1.09% and respectively 1.98% higher than the accuracy obtained by learning algorithms alone. Additionally, the reduction of running time especially for time consuming learning algorithms like SVMO is highly significant, 59% and respectively 91%.

Full Document (PS)

Bibtex

@TechReport{csfsadm,
    author = "Lumini{c t}a Mihaela Moruz and Liviu Ciortuz",
    title = "A Comparative Study on Feature Selection Algorithms in Data Mining",
    institution = "``Al.I.Cuza'' University of Ia{c s}i, Faculty of Computer Science",
    year = "2005",
    number = "TR 05-06",
    url = "https://publications.info.uaic.ro/technical-reports/archive/tr05-06-2005-a-comparative-study-on-feature-selection-algorithms-in-data-mining/"
}