In the last decade, airborne light detection and ranging (LiDAR) scanning (ALS) technology has become a powerful technique for remote sensing, imaging, and mapping. However, the data obtained from any measurement system can include inaccurate signals affected by systematic errors or by the external environment. De-noising to remove inaccurate outlier points is a fundamental and challenging problem for ALS-based mapping applications. The proposed method aims to recover the patterned (planar and linear) points within the assigned outlier and removed points. The method consists of 3 steps. First, statistical outlier removal (SOR) filtering is implemented, and outlier points are detected with the filtering method. Next, the machine learning system reclassifies the filtered outlier points. If the classification result is “inlier” , that point is added to the filtered inlier point cloud as an inlier point. The accuracy of outlier points was evaluated against a manually determined validation set. The results achieved 99 % and 98 % according to the highest overall accuracy criterion and kappa coefficient, respectively. These findings are a promising step to test the proposed method in three different test areas and extend it to widespread spatial dimensions. Furthermore, the findings show that many useful points are removed by SOR filtering. The developed methodology contributes to the reduction of errors caused by data losses in various modelling studies, especially for power transmission line and 3D façade modelling studies.