The classification of the urban point cloud is an essential task for numerous applications, including mapping, 3D urban modelling, etc.. Although in the last few years, different methodologies and algorithms have been proposed, precise and detailed point cloud labelling is still challenging. Publicly available annotated benchmark datasets have become the standard for the evaluation of algorithms' performance; however, most focus on data acquired from mobile or terrestrial laser scanners. In this paper, we introduce UNS Geo, a dense Aerial Laser Scanning (ALS) point cloud dataset consisting of 5.4 million manually annotated points across 8 semantic classes. To validate the performance of our dataset, the labelled point cloud is used for training the state-of-the-art networks (i.e. PointNet, PointNet++). Moreover, since UNS Geo includes the RGB per point information, the influence of spectral information on classification results is evaluated. The results demonstrate that UNS Geo effectively supports the training of deep learning models, highlighting its potential for advancing research in urban point cloud classification. The dataset is publicly available at: https://github.com/mirogovedarica/UNS-Geo.