This work investigates the use of machine learning to predict the elastic stress concentration factor based on 3D scans of 135° V-notched round bar specimens. Traditionally, fatigue assessment of notched components relies on theoretical values. However, is highly dependent on the actual notch geometry, which can deviate significantly from the idealized geometry. These deviations impact the severity of the notch stress concentration and, consequently, the fatigue lifetime of a component. Therefore, fatigue assessment based on real notch geometries is of great interest. The experimental work involved detailed 3D scanning of V-notched S690 steel specimens using a Keyence VR-5200 profilometer. These measurements were crucial for developing 151 finite element models, incorporating the actual notch profile data. In addition, the influence of specimen manufacturing on the experimental fatigue life was investigated by conducting rotating bending fatigue experiments on both polished and non-polished specimens. The scans of the V-notched specimens were post-processed to create three-dimensional solid finite element models. For each of these finite element models, the corresponding -value was determined. These results served as the training database for the machine learning models. The input to the machine learning models is post-processed geometrical data derived from the 3D scans and the target variable is the corresponding -value determined with finite element analysis. Various machine learning models were evaluated, with gradient boosting regression demonstrating the highest performance, achieving a mean absolute error of 0.039 and a coefficient of determination of 0.909 using features selected through recursive feature elimination. Careful preprocessing, resampling and data augmentation were critical for developing high-performance models. A case study highlighted the benefit of using profile data over discrete geometrical notch parameters when evaluating the stress concentration factor.