Power converter faults in wind turbines often lead to costly downtime and repeated maintenance. We present a practical, explainable, and fully data-driven approach that utilizes high-resolution converter control system records, 1 min operating data, and event logs to predict whether a fault leads to a long or short standstill. By combining engineered features with interpretable feature reduction, we achieve 89 % accuracy and an F1 score of 0.86, providing support for remote decision-making.