Reliable depression prediction based on deep learning is crucial for clinical applications, requiring both prediction reliability and algorithmic fairness across diverse demographic groups. In this study, we investigate the algorithmic fairness of uncertainty quantification (UQ), namely Equal Opportunity Coverage (EOC) fairness, and propose Fair Uncertainty Quantification (FUQ) for depression prediction. FUQ pursues reliable and fair depression prediction through group-based analysis. Specifically, it groups participants by different sensitive attributes and quantifies uncertainty within each demographic group using conformal prediction. Furthermore, we propose a fairness-aware optimization strategy that formulates fairness as a constrained optimization problem under EOC constraints. The proposed method is demonstrated to be effective through extensive evaluation on multiple visual and audio depression datasets.
