A Neural Gas Based Approximate Spectral Clustering Ensemble

Moazzen Y., TAŞDEMİR K.

11th International Workshop on Self-Organizing Maps (WSOM), Texas, Amerika Birleşik Devletleri, 6 - 08 Ocak 2016, cilt.428, ss.85-93 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 428
  • Doi Numarası: 10.1007/978-3-319-28518-4_7
  • Basıldığı Şehir: Texas
  • Basıldığı Ülke: Amerika Birleşik Devletleri
  • Sayfa Sayıları: ss.85-93
  • İstanbul Teknik Üniversitesi Adresli: Evet

Özet

The neural gas has been successfully used for prototype based clustering approaches. Its topology based quantization effectively aids in approximate spectral clustering (ASC) to define distinct similarity criteria which are optimally selected for the relevant application. To utilize the advantages of ASC by harnessing those criteria derived from different information types, we propose a neural gas based approximate spectral clustering ensemble (NGASCE). The NGASCE obtains a joint decision for accurate partitioning, by a 2-step ensemble approach derived from 1-step graph-based models. We show the outperformance of NGASCE on five datasets from UCI Machine Learning Repository.