The reliable assessment of the quality states of machine tools offers significant opportunities for optimizing existing production chains and planning maintenance work. In particular, the domain of radial-axial ring rolling, which is a hot forming process to produce seamless rings, offers significant potential for optimization. Previous research work has shown that information on the quality status of forming rolls is currently limited and that data-driven approaches such as machine learning processes can help to solve this problem. Accordingly, this paper uses a random forest time series classification approach to evaluate process data on their informative value for the mandrel quality assessment. The study demonstrated that the individual process features in ring rolling are of limited value in providing information about the condition of the mandrel. Only in combination with other features, which contain information about the previous rolling of the mandrel roll, can meaningful results be obtained.