Stereoscopic images and videos can lead to serious adverse effects on human visual perception. The phenomenon of visual discomfort depends on various influencing factors such as the arrangement of the display system, the image quality and the design of 3D effects. Real-time depth adaptations that reduce the extent of visual discomfort require computationally efficient prediction models. This article analyzes optimal combinations of image features of state-of-the-art models in terms of prediction accuracy and computational efficiency. In addition, a fast-to-compute disparity contrast feature based on Haralick contrast is introduced in this context. It turns out that the computational complexity can be reduced by restricting the number of features without loss of prediction accuracy. A Pareto-front analysis shows which features are more likely to be part of optimal combinations. It is interesting to observe that the introduced disparity contrast feature is part of combinations that are optimal in terms of both computational efficiency and accuracy. This means that state-of-the-art prediction models can be improved by means of the introduced disparity contrast feature. The analysis relies on statistical evaluations based on publicly available assessment data.