Robust mixture modeling using the truncated t distribution: recent advances and new results

The truncated t (TT) distribution has been recognized as a flexible and powerful framework for handling truncated and heavy-tailed characteristics. This paper is devoted to an in-depth investigation of its mathematical properties and extends its applicability to a finite mixture formulation, referred to as the FM-TT model henceforth. An analytical Expectation Conditional Maximization Either (ECME) algorithm is developed for parameter estimation. A key advantage of this approach is that the E-step is carried out analytically, thereby preventing the need for computationally intensive Monte Carlo integration. Furthermore, a general information matrix-based approach is exploited to assess the variability of the estimated parameters. The practical utility of the proposed methodology is demonstrated through extensive experiments on both simulated and real-world datasets. Experimental results validate the effectiveness of the FM-TT model in capturing data structure and its superiority over existing approaches by achieving better model fits and improved classification accuracy.