A new partial robust adaptive modified maximum likelihood estimator

Acitas S., Filzmoser P., ŞENOĞLU B.

CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS, cilt.204, 2020 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 204
  • Basım Tarihi: 2020
  • Doi Numarası: 10.1016/j.chemolab.2020.104068
  • Dergi Adı: CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Applied Science & Technology Source, Biotechnology Research Abstracts, Chemical Abstracts Core, Chimica, Computer & Applied Sciences, EMBASE, INSPEC
  • Anahtar Kelimeler: Partial least squares, Robust adaptive modified likelihood, Outliers, Robust estimation, PARTIAL LEAST-SQUARES, REGRESSION, PARAMETERS
  • Ankara Üniversitesi Adresli: Evet

Özet

Partial least squares (PLS) regression is a widely-used regression method for high-dimensional data. However, PLS is not robust to outlying observations since it uses partial information of the variables in a least squares (LS) setting, which is known to be very sensitive to outliers. Several proposals are available which robustify PLS. In this study, our aim is to propose a new partial robust estimator using robust adaptive modified maximum likelihood (RAMML) estimators [1, 2]. The resulting estimators are therefore called partial robust adaptive modified maximum likelihood estimators (PRAMMLs). The distinguished advantage of the PRAMMLs is that they are computationally straightforward. This is because of the fact that they are constructed based on explicitly formulated estimators. The simulation study shows that the PRAMMLs are preferable to PLS and other existing robust alternatives of PLS in terms of the mean squared error (MSE) criterion under different nonnormal error distributions, as well as in the presence of leverage points. The PRAMMLs also give satisfactory results in terms of the empirical influence function and breakdown robustness criteria.