<p>In modern industrial processes, the availability of labeled anomaly data can hardly be found, which makes the application of traditional supervised fault detection models ineffective. To address this issue, this paper provides an elaborated framework of anomaly detection based on an optimized One-Class Support Vector Machine (OC-SVM). An electrowinning process simulation was used to create a labeled multivariate dataset (Voltage, Current, and Temperature) in both normal operation conditions and three different fault conditions short circuits, electrolyte imbalances and heating failures. OC-SVM was only trained on standard operational data and hyperparameters were optimized systematically using a grid search strategy to ensure that the detection performance is maximized. The optimized model showed outstanding performance on the hidden test data, with the ideal Area Under the ROC Curve (AUC) was 1.000, and the recall of the anomaly was 1.000, to make sure there were no instances of missed faults. The model had an anomaly F1-score of 0.990 and just three false positives were obtained. Moreover, the visualization of the decision boundary in the form of geometric shapes showed that the model is robust because the normal cluster of data was accurately enclosed. The given results confirm the effectiveness of the suggested one-class learning model as the efficient solution that can be used to monitor the processes in the industry, where the data distribution is highly skewed.</p>