Simultaneous Feature Selection and ANN Training for Data Mining Applications

Abstract: Feature selection is the method to find which features to be used for successful predictions. In a given dataset to be predicted there can be some inputs that are not important and do not affect the result much. These inputs can be eliminated for simpler and faster prediction. After elimination, the model gives very close results compared to the first model. It may even make the results better and more successful on some datasets. Feature selection has some advantages. Firstly it provides faster prediction model, improves the prediction success, and finally helps to understand relation of inputs and output easier. Artificial neural networks (ANNs) can be used for classification and they have very high accuracy rates. In the training process of ANNs, feature selection can provide advantages. For applying feature selection, it is necessary to decide which inputs to be eliminated. The elimination can be made by using a meta-heuristic algorithm. In this study, a new algorithm is proposed for simultaneously training ANNs and feature selection. For this purpose ant colony optimization (ACO) algorithm is used for feature selection and for each ant of the population at each iteration feed-forward ANNs are trained by using stochastic fractal search of the ACO algorithm. The obtained results of the proposed algorithm are compared with the full-input model on some well known classification benchmark problems and promising results are obtained.

Keywords: Feature Selection, Artificial Neural Network, Ant Colony Optimization, Stochastic Fractal Search

Publication details: Conference abstract presented at the 4th International Conference on Engineering and Natural Sciences (ICENS 2018), 2-6 May 2018, Kyiv, Ukraine. Published in: Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018), pp. 220. Publisher: ICENS. ISBN: 978-605-67955-2-7.

How to cite / Kaynak gösterimi

APA 7: Gülmez, B., & Kulluk, S. (2018, May). Simultaneous Feature Selection and ANN Training for Data Mining Applications [Conference abstract]. In Ö. Çınar (Ed.), Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018) (p. 220). ICENS. 4th International Conference on Engineering and Natural Sciences (ICENS 2018), Kyiv, Ukraine. https://doi.org/10.5281/zenodo.23036887

IEEE: B. Gülmez and S. Kulluk, "Simultaneous Feature Selection and ANN Training for Data Mining Applications," in Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018), Kyiv, Ukraine, May 2018, p. 220, ISBN: 978-605-67955-2-7, doi: 10.5281/zenodo.23036887.

Chicago: Gülmez, Burak, and Sinem Kulluk. 2018. "Simultaneous Feature Selection and ANN Training for Data Mining Applications." In Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018), 220. Kyiv, Ukraine: ICENS. https://doi.org/10.5281/zenodo.23036887.

MLA 9: Gülmez, Burak, and Sinem Kulluk. "Simultaneous Feature Selection and ANN Training for Data Mining Applications." Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018), ICENS, 2018, p. 220. https://doi.org/10.5281/zenodo.23036887.

Harvard: Gülmez, B. and Kulluk, S. (2018) 'Simultaneous Feature Selection and ANN Training for Data Mining Applications', in Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018). Kyiv, Ukraine: ICENS, pp. 220. doi: 10.5281/zenodo.23036887.

BibTeX:

@inproceedings{Gulmez2018simultaneous,
  author    = {Gülmez, Burak and Kulluk, Sinem},
  title     = {Simultaneous Feature Selection and ANN Training for Data Mining Applications},
  booktitle = {Book of Abstracts of the IV International Conference on Engineering and Natural Sciences (ICENS 2018)},
  year      = {2018},
  month     = {may},
  pages     = {220},
  address   = {Kyiv, Ukraine},
  publisher = {ICENS},
  isbn      = {978-605-67955-2-7},
  doi       = {10.5281/zenodo.23036887},
  url       = {https://doi.org/10.5281/zenodo.23036887}
}