3.3.2. Domain-Specific Methods

This method uses the knowledge and experience in the domain from which the data originates. Domain-specialization and manual treatment of the data are required. This can be both expensive and time consuming but the end-result, i.e., the cleaned data, is expected to be of higher quality than using pure statistical outlier detection algorithms, which do not adhere to the domain-specific considerations. In the scope of the EAF process, an example of a domain-specific treatment is the removal of instances of data that are improbable. If the registered charged content of scrap is 190 t while the maximum capacity of the furnace is 120 t, then that instance should be removed. Another example is the removal of heats that are not part of regular production. For example, testing the effects of new scrap types, or delivery batches.
