Predicting the next sample in a correctly working system with an auto-regressive model
Since we only have normal values in the training set, we train and test an auto-regressive model to predict normal values. Then, during deployment, we set an alarm system based on the distance calculated between the real values and predicted values. The concept is that the auto-regressive model can only predict values reflecting a correctly functioning system. If the underlying functioning system starts deteriorating, then the predicted values and the real values will start diverging.
In the following steps, we will introduce the auto-regressive model approach and our implementation:
- Introducing an auto-regressive model
- Training an auto-regressive model with the linear regression algorithm
- Deploying an auto-regressive model
In the first subsection, we’ll introduce the linear regression algorithm as one option for training an auto-regressive model.
