Supervised Learning

Training stage

1. A dataset with input data and the corresponding output.

2. The input data is pre-processed to identify and/or extract relevant features.

3. The feature data is input to the ML method, typically one feature set (e.g: mean, median, std, etc).

   1. Sometimes a blind set of features is produced, and then only the most relevant are selected (e.g: decision trees).

4. For each input set, the method produces an estimated likeliness of an error occurrence.

5. The method compares the estimate with the actual process output (the Ground Truth) and updates the model's internal processes to improve the accuracy of the estimates.

6. The process is replicated until the performance of the method is within acceptable bounds.

Inference stage

1. The trained model is deployed in its target setting.

2. Given inputs, it can produce estimates of the process output.

3. However, the method no longer has access to the ground truth and is thus unable to further its learning.