<p>Graph Neural Networks (GNNs) have achieved notable success in a wide range of graph-based learning applications. However, they still face critical challenges such as over-smoothing and limited robustness. This work introduces a unified framework designed to concurrently address these challenges through adaptive regularization and multi-objective optimization. Our approach proposes an adaptive DropEdge strategy that dynamically modulates edge dropout rates based on both structural properties and node feature distributions, thereby preserving essential graph information. Furthermore, we formulate a multi-objective optimization problem that jointly minimizes a classification loss and a consistency regularization loss, encouraging both accurate predictions and stable representations across graph perturbations. This problem is efficiently solved using NSGA-II, an evolutionary algorithm well-suited for balancing competing objectives. Experimental results on several widely used benchmark datasets demonstrate that our approach not only mitigates over-smoothing but also improves robustness, providing a more effective and scalable solution for GNNs.</p>