A New Dual-phase Hybrid Variable Selection Method in High-Dimensional Data

Sancar N., Arkanöz B., İNAN D.

8th International conference Approximation Theory and Special Functions, ATSF 2024, Ankara, Türkiye, 4 - 07 Eylül 2024, cilt.503 PROMS, ss.301-319, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 503 PROMS
  • Doi Numarası: 10.1007/978-3-031-93279-3_16
  • Basıldığı Şehir: Ankara
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.301-319
  • Anahtar Kelimeler: Elastic-net, High-dimensional data, Logistic model, Particle swarm optimization, Variable selection
  • Marmara Üniversitesi Adresli: Evet

Özet

This study aims to propose a dual-phase hybrid variable selection in high-dimensional data for the logistic model. In the first phase of the proposed method, the modified version of the binary particle swarm optimization (BPSO) approach as the wrapper method was utilized to lower the number of variables because of its numerous advantages. Like other metaheuristic algorithms, BPSO cannot guarantee finding the optimal solution. To address this issue, the Elastic Net (EN) method was employed as the embedded approach in the second phase to enhance the variable selection process due to its advantages over other embedded techniques. The variable selection efficiency of the proposed method was compared with the Elastic-net regularization method under various simulated high-dimensional data and real data using various evaluation metrics. The findings indicate that the proposed BPSO-Elastic Net method is the highly reliable method for variable selection in high-dimensional, where it outperforms the traditional Elastic Net method.