Mixed-mode fracture behavior of metal materials like steel is often assessed with Compact Tension Shear Specimen proposed by Richard [1]. This specimen enables testing fracture behavior under fatigue loading with different mixicity ratio, which can be regulated by changing the loading angle. This approach is commonly used, but also have its own challenges. To properly asses the results researchers have to find the stress intensity factors for mode I and mode II for the points on the growing crack. However analytical solutions provided by Richard are only valid for the starting geometry (with initial straight crack). Therefore, numerical calculations has to be provided (commonly used solution here is finite element analysis). This, however, can be time-consuming process. In the present times, one of the possible solutions to this type of the challenges is to use machine learning approaches. As shown in [2] it is possible to use mixed numerial – machine learning approach to predict mechanical quantities.