Evaluation Metrics

Model evaluation is indispensable for creating effective models that not only perform well on the data that was used to train the model but also on unseen data. The task of evaluating the model is especially easy when dealing with supervised learning problems, where there is a ground truth that can be compared against the prediction of the model.

Determining the accuracy percentage of the model is crucial for its application to unseen data that does not have a label class to compare to. For example, a model with an accuracy of 98% may allow the user to assume that the odds of having an accurate prediction are high, and hence the model should be trusted.

The evaluation of performance, as mentioned previously, should be done on the validation set (dev set) to fine-tune the model, and on the test set to determine the expected performance of the selected model on unseen data.

Evaluation Metrics for Classification Tasks

A classification task refers to a model...