Large-scale 3D point clouds are essential for autonomous driving but suffer from significant class imbalance, where certain object categories dominate while others are underrepresented. This imbalance often causes the model to be biased toward classes with the most data or distinct physical shapes. Consequently, when underrepresented classes are surrounded by majority classes, the model becomes confused (e.g., road markings adjacent to the ground) and performs poorly. To address this fairness issue, we propose a new approach that is an enhanced version of RandLA-Net using an ensemble method, AdaBoost, which adaptively reviews misclassified samples to improve performance for underrepresented classes overshadowed by majority classes. Then, we employ a fairness-aware evaluation using a weighted Intersection over Union ( wIoU) metric to emphasize underrepresented classes, ensuring a fair evaluation of segmentation performance across all classes. Experimental results demonstrate that the RandLA-NetBoosting outperforms the original RandLA-Net, mitigating class imbalance bias and improving overall segmentation accuracy. This illustrates the potential of fairness-aware methods to address real-world challenges in autonomous driving.