Selection Bias

Selection bias exists when the samples collected are not representative of the entire group. Thus, random selection bias occurs when the data used to train or evaluate an AI model is not representative of the part variation due to an erroneous or non-random selection procedure. This leads to AI models that perform well with the selected data, but poorly with real data. For example, if an AI model for predicting part defects is only trained with data from a particular part, it may not be possible to apply it to other parts with different defect patterns, which raises the question: “Is the training data of the AI model really representative of the wider part variation it will face?”