A Novel Regress Information Criterion for Gaussian Mixture Model
Shuping Sun, Shengmei Mo, Yizhuo Zhang, Guangyu Liu, Jinbo Chen, Dasheng Liu, Yaonan Tong · 2025
Aimed at what criterion employed to determine the optimal number of components when using a Gaussian mixture model (GMM) to adaptively fit a given data set ($X$) without any prior knowledge, this study proposes a novel synthesized criteria (SYC)-based regress information criterion (RIC) to determine an optimal number of components ($M^{\text {RIC }}$). The performance evaluation of RIC and its robust characteristics were verified by comparing with other criteria on the simulated and real data sets. The comparative results indicate that the RIC can replace SYC and synthesize the advantages of the Akaike information criterion (AIC) applied for smaller sample sizes and the Bayesian information criterion (BIC) for larger sample sizes.