In confusion matrix prediction, has the training data used for prediction been seen by the model

Question

I understand that Confusion Matrix evaluates prediction results with metrics such as precision, recall, accuracy and F1-Score.

My question is regarding the data used for prediction. Is it okay if this data has been seen by the model? That is, if the data used for generating predictions was the same data used for training the model, is it still valid to use confusion matrix to evaluate the model performance? Or should I used completely new data where the model has never seen for making predictions?

I want to learn about the model performance but before that I want to confirm what data the predictions are made based on so that the it is valid to use confusion matrix for model performance evaluation.

Answer 1

The predictions are made using the data that you give to the model to make predictions. The model predictions are informed by the training data. If you’re trying to assess how the model performs on unseen data, you need to give it unseen data.