This paper presents a framework for systematically estimating uncertainty quantification (UQ) to assess the reliability of predictions in probabilistic machine learning models. Specifically, we focus on the Gaussian Process Latent Variable Model (GPLVM), which efficiently approximates the predictive distribution using a scalable Random Fourier Feature-based Gaussian Process. This model estimates both epistemic and random uncertainty, derives a theoretical formulation for UQ, and proposes a Monte Carlo sampling-based estimation method. Experiments demonstrate the impact of uncertainty estimation, provide insights into the sources of predictive uncertainty, and demonstrate the effectiveness of the proposed approach.
