Feature Selection with Artificial Bee Colony Algorithms for Classifying Parkinson’s Diseases


Durgut R. , Baydilli Y. Y.

21st International Conference on Engineering Applications of Neural Networks, Halkida, Yunanistan, 4 - 06 Haziran 2020, ss.338-351

  • Basıldığı Şehir: Halkida
  • Basıldığı Ülke: Yunanistan
  • Sayfa Sayıları: ss.338-351

Özet

Parkinson’s is a brain disease that affects the quality of human life significantly with very slow progresses. It is known that early diagnosis is of great importance to arrange relevant and efficient treatments. Data analytics and particularly predictive approaches such as machine learning techniques can be efficiently used for earlier diagonosis. As a typical big data problem, the number of features in the collected data of Parkinson’s symptoms per case matters crucially. It is known that the higher the number of features considered the more complexities incur in the handling algorithms. This leads to the dimensionality problem of datasets, which requires optimisation to overcome the trade-off between complexity and accuracy. In this study, artificial bee colony-based feature selection methods are employed in order to select the most prominent features for successful Parkinson’s Disease classification over the datasets. The optimised set of features were used in training and testing k nearest neigbourhood algorithm, and then verifed with support vector machine algorithm over the public dataset. This study demonstrates that binary versions of artificial bee colony algorithms can be significanlty successful in feature selection in comparison to the relevant literature.

Parkinson’s is a brain disease that affects the quality of human life significantly with very slow progresses. It is known that early diagnosis is of great importance to arrange relevant and efficient treatments. Data analytics and particularly predictive approaches such as machine learning techniques can be efficiently used for earlier diagonosis. As a typical big data problem, the number of features in the collected data of Parkinson’s symptoms per case matters crucially. It is known that the higher the number of features considered the more complexities incur in the handling algorithms. This leads to the dimensionality problem of datasets, which requires optimisation to overcome the trade-off between complexity and accuracy. In this study, artificial bee colony-based feature selection methods are employed in order to select the most prominent features for successful Parkinson’s Disease classification over the datasets. The optimised set of features were used in training and testing k nearest neigbourhood algorithm, and then verifed with support vector machine algorithm over the public dataset. This study demonstrates that binary versions of artificial bee colony algorithms can be significanlty successful in feature selection in comparison to the relevant literature.