The mixture of experts framework is widely utilized in statistics and machine learning to address data heterogeneity in tasks such as regression, classification, and clustering. In clustering continuous data, the mixture of experts typically employs experts that follow a Gaussian distribution. However, outliers can adversely affect clustering outcomes. To address this issue, various methods have been proposed in the literature. In this paper, we introduce a novel approach that models the experts using the symmetric α-stable distribution. This flexible distribution effectively accommodates different types of outliers (especially extreme outliers) and skewness, while also encompassing Gaussian experts as a special case when α =2. The maximum likelihood estimates of the model parameters (excluding α) are obtained using an expectation-maximization approach, while α is estimated using Monte Carlo integration and interpolation. The effectiveness of this approach is demonstrated through analyses of both real and simulated data.
Zarei, S. (2025). Robust mixture of experts modeling using symmetric α-stable distributions. Journal of Statistical Modelling: Theory and Applications, 6(1), 133-150. doi: 10.22034/jsmta.2026.23048.1180
MLA
Zarei, S. . "Robust mixture of experts modeling using symmetric α-stable distributions", Journal of Statistical Modelling: Theory and Applications, 6, 1, 2025, 133-150. doi: 10.22034/jsmta.2026.23048.1180
HARVARD
Zarei, S. (2025). 'Robust mixture of experts modeling using symmetric α-stable distributions', Journal of Statistical Modelling: Theory and Applications, 6(1), pp. 133-150. doi: 10.22034/jsmta.2026.23048.1180
CHICAGO
S. Zarei, "Robust mixture of experts modeling using symmetric α-stable distributions," Journal of Statistical Modelling: Theory and Applications, 6 1 (2025): 133-150, doi: 10.22034/jsmta.2026.23048.1180
VANCOUVER
Zarei, S. Robust mixture of experts modeling using symmetric α-stable distributions. Journal of Statistical Modelling: Theory and Applications, 2025; 6(1): 133-150. doi: 10.22034/jsmta.2026.23048.1180
)