<p>Recording dance movements nowadays becomes problematic due to complex recording procedures and unavoidable data loss caused by some resource elements, like bodily or clothing material composition. The task of filling in the missing data for the performed motion and retrieving the sequence as a whole becomes difficult due to the characteristics of physical motion, which include cinematographic perspectives that render the movements themselves non-linear. Previous works have indicated some level of success in loss motion recovery, but only for a short span. The first two-dimensional matrix computation paradigm lacks theoretical justification for the recovery of the non-linear motion information, which is a limitation. This issue has been addressed by developing a new enhanced model called the Machine Learning 2-Dimensional Matrix-Calculation (ML-2DMC), which is presumably designed to achieve the rehabilitation and recovery of human movement and dance. The proposed procedure takes advantage of the effectiveness of the machine learning algorithms and applies 2D matrix computation methods, permitting good results across a variety of experiments. A new method called fractal-chaotic map grey wolf optimizer (FCM-GWO) is introduced to optimize the parameters of ML-2DMC. This optimization itself increases the efficiency of the ML-2DMC model when it comes to the retrieval of complex movements of the processes involving dance. The paper gives experimental results validating the efficiency of the proposed approach against other methods, such as recurrent convolutional neural networks and other more sophisticated models and approaches incorporating multi-paradigm sensors and devices such as Kinect sensors along with low-rank matrix completion methods. The study shows that the ML-2DMC-FCM-GWO method effectively tackles the complexities of non-linear human motion and dance recovery, making a significant addition to the field of motion analysis and restoration.</p>