A Dataset Fusion Algorithm for Generalised Anomaly Detection in Homogeneous Periodic Time Series Datasets
The Dataset Fusion algorithm merges multiple, different but similar time series datasets into one unified set for training an unsupervised LSTMCaps Neural Network to detect anomalies. Tested on two induction motor fault datasets, this approach achieved a high average F1 score of 0.879 and shows strong generalization across all data, proving its effectiveness in creating robust anomaly detection models.