Nordic Edge AI Lab enables you to train anomaly detection models in an unsupervised manner. You do not need to label your data as normal or abnormal. Simply upload a dataset representing normal operational sensor data, and the system learns to recognize it as the normal mode of operation. During inference, if new sensor data falls outside these learned patterns, it is classified as abnormal.
Unlike classification or regression, the anomaly detection task in Neuton always trains a single model and does not support adding a holdout validation dataset. Model validation is performed by running inference on datasets with and without anomalies and analyzing the predicted anomaly scores.