Abstract-In recent years, one-class support vector machine (SVM) approaches have received particular attention in fault detection since only one class of the data is required for training. However, the training data can be corrupted with the outliers that influence classifier performance significantly. In this paper, a Gaussian-based penalisation has been proposed in the formation of a robust one-class SVM model which constructs the decision boundaries that are robust to the outliers without compromising the classification performance. The efficacy of the proposed method has been compared with the traditional one-class SVM and a previous robust one-class SVM method in the literature when applied in three datasets: the Iris's Fisher dataset, banana-shaped dataset and MFPT bearing fault dataset. It is shown that the proposed robust one-class SVM outperforms other methods.