In the case of natural public objects such as slopes, it is difficult even for specialist engineers to detect anomalies early due to the significant influence of uncertainty in deterioration. Under such circumstances, it is effective to detect anomalies on slopes using spatially high-density point cloud data. In civil engineering, laser measurement technology used for point cloud data measurement represented by the Mobile Mapping System (MMS) has been developed. However, point cloud data are not yet fully utilized for decision-making in practice as it is not possible to easily grasp the deformation and changes of each feature. In this study, the authors propose a local anomaly detection method for slopes based on point-cloud deep learning with point cloud data as input. Furthermore, we make attempts to detect the frame blocks with local anomalies such as deformation and protrusion, and empirically analyze the effectiveness of the method proposed in this study.