How does this work technically?
Anomaly prediction is the detection and prediction of exceptional events, anomalies. Compared to label prediction, and certainly forecasting, anomalies are events that occur even less frequently, such as the occurrence of a hurricane. Because these events occur so rarely, there is often not much information in the training data on which the algorithm can rely to detect relationships when these anomalies do or do not occur. In other words, there is an imbalance in the training data. As a result, other types of algorithms, and data processing, are required. In the hypertuning process of anomaly prediction, techniques such as SMOTE, undersampling or oversampling are frequently used. The entire hypertuning process is automated just like with all other AI and Deep Learning applications within Trendskout.