Table 6.

Monitoring cost related to the classification decision of the ML model, and used as one of the criteria for ML selection.

Prediction result	Definition of the prediction result	Monitoring
Prediction result	Definition of the prediction result	Follow-up actions	Costs components per batch^a	Explanation of the follow-up actions
True Positive	Non-compliant batch predicted to be non-compliant	Sampling & Analysis	C_coll + C_analy + C_stor	Sampling & Analysis and holding, followed by rejection of the non-compliant feed batch to feed material traders
		Storage
False Positive	Compliant batch predicted to be non-compliant	Sampling & Analysis	C_coll + C_analy + C_stor	Sampling & Analysis and holding, followed by acceptance of the compliant batch
		Storage
True Negative	Compliant batch predicted to be compliant	Accept	0	Accept compliant batch
False Negative	Non-compliant batch predicted to be compliant	Accept, Recall, Destroy, Replace, Estimate disease burden	P_recall * (C_recall+ C_destr+ C_price)10+(1 − P_recall) C_burden	1. Accept based on prediction, or 2. Recall and other actions due to non-compliant feed batch that is used to produce compound feed and found to be contaminated later

^aSee Table 7 for explanation of the cost component variables.