Autors: Naka, E., Guliashki, V. G., Marinova, G. I.
Title: A Comparative Analysis of Different Feature Selection Methods on Parkinson Data
Keywords: Random Forest, k-nearest neighbour, Parkinson dataset

Abstract: The paper provides a comparison of different feature selection methods for a voice Parkinson dataset in order to find an optimal subset with relevant features which gives the higher accuracy. Preliminary, the search space can be reduced using different methods toward searching for relevant features and to decrease the time complexity of the algorithm. Different filter and wrapper methods are applied to this dataset. The performance of each feature selection method is evaluated through the accuracy of three classifiers: Support Vector Machine specifically for the radial kernel basis function, Random Forest, and k-nearest neighbour. Moreover, an optimization algorithm Generalized Simulated Annealing is applied for hyper-parameter tuning of each classifier with the objective of improving its accuracy. Another algorithm, the Genetic Algorithm has been also applied for finding the best optimal subset of features and again is tested with the above-mentioned classifiers.

References

    Issue

    15th International Conference on Advanced Technologies, Systems and Services in Telecommunications, TELSIKS’2021, pp. 366-371, 2021, Serbia, IEEE, doi: 10.1109/TELSIKS52058.2021.9606398.

    Вид: публикация в международен форум, публикация в реферирано издание, индексирана в Web of Science