@phdthesis{digilib79783,
           month = {August},
           title = {ANALISIS DISKRIMINAN KUADRATIK ROBUST DENGAN ESTIMATOR MINIMUM COVARIANCE DETERMINANT (MCD)},
          school = {UIN SUNAN KALIJAGA YOGYAKARTA},
          author = {NIM.: 22106010083 Fernanda Devita Sukma},
            year = {2026},
            note = {Dr. Epha Diana Supandi, S.Si., M.Sc.},
        keywords = {Analisis Diskriminan Kuadratik Robust, Klasifikasi Kemiskinan,
Minimum Covariance Determinant, Outlier},
             url = {https://digilib.uin-suka.ac.id/id/eprint/79783/},
        abstract = {Disparities in poverty levels among regencies/cities in East Java Province require a
classification method capable of accurately grouping regions to support the
formulation of appropriately targeted policies. Quadratic Discriminant Analysis
(QDA) is a multivariate classification method that accommodates differences in
covariance matrices among groups; however, its parameter estimation is sensitive
to the presence of outliers. This study aims to compare the classification
performance of classical QDA and robust QDA using the Minimum Covariance
Determinant (MCD) estimator in classifying the poverty levels of 38
regencies/cities in East Java Province in 2024 based on five socioeconomic
indicators. The analysis consisted of multivariate normality testing using the
Henze?Zirkler test, outlier detection using Mahalanobis distance, construction of
classical QDA and QDA Robust-MCD models, and model evaluation using the
Apparent Error Rate (APER) and classification accuracy. The results showed that
the data did not satisfy the multivariate normality assumption, and three outlier
observations were identified among the 38 regencies/cities analyzed. Classical
QDA achieved a classification accuracy of 78.95\% with an APER of 0.2105,
whereas QDA Robust-MCD achieved an accuracy of 71.05\% with an APER of
0.2895. The comparison results indicate that classical QDA provided better
classification performance than QDA Robust-MCD based on both classification
accuracy and APER. Thus, for the data analyzed in this study, the use of the MCD
estimator did not result in improved classification performance compared with
classical QDA.}
}