A Bayesian estimation approach for the wavelet-based mixture regression

Most methods in mixture modeling assume that the mixture weights have a constant behavior. In practice, this assumption restricts the applicability on data sets where weights are expected to be time- or space-dependent. In this article, we relax this assumption and propose a method to estimate dynamic mixture weights within a two-component Gaussian mixture model. Unlike existing approaches, our method leverages Bayesian inference to estimate the mixture component parameters. The proposed method accurately captures the dynamics of the mixture weights across various simulated scenarios. Moreover, when applied to real data from tumor glioblastoma multiforme (GBM) studies, our method reliably identified probability peaks associated with chromosomal abnormalities linked to GBM presence.