Anomaly detection is a crucial aspect of embedded applications. However, limited computational power, evolving environments or lack of training data are difficulties that can limit anomaly detection algorithms. Anomaly detection can be performed by one class classification algorithms to remove the need for anomalous data in the training set. This paper presents a new machine learning algorithm for anomaly detection called Dynamic Double anomaly Detection Dyd². A thorough description of Dyd² is performed. Then an experimental evaluation is set up to compare Dyd² to state-of-the-art algorithms.