Generative models have achieved remarkable success in state-of-the-art image and text tasks. Recently, score-based diffusion models have extended their success beyond image generation, showing competitive performance with discriminative methods in image classification tasks (Zimmermann et al., 2021). However, their application to classification in the graph domain, which presents unique challenges such as complex topologies, remains underexplored. In this work, we propose a modified training objective to enhance graph classification performance using score-based diffusion models, while preserving their generative capabilities. Additionally, recognizing the resource-intensive nature of these models during classification inference, we introduce an efficient model selection method based on the validation set loss.