How do I choose the most suitable evaluation metric for my classification model?

Evaluation criteria for classification models include accuracy, precision, recall, and F1 score. How can I choose the evaluation metric that is most appropriate for my dataset and problem?
As an illustration, I’ve identified all the metrics values in the code below, but I’m unable to determine which is the best option for my issue or determine whether the model is well fitted to my data.


1- The overall correctness of the predictions is measured by accuracy. The assumption that high accuracy equates to good modeling is a typical error. However, accuracy can be deceptive, particularly in datasets that are unbalanced. A classifier that predicts only the prevailing class could have high accuracy without being useful if one class dominates the other. Although accuracy is crucial, it might not be acceptable for datasets that are unbalanced. When classes are unequally distributed, metrics like precision, recall, and F1 score offer a more complete picture of the model's performance.

2- How many of the predicted positive cases are positive is known as precision. Beginners might just prioritize improving precision without considering the trade-off between recall and precision. When a recall is sacrificed, maximizing precision could result in missing significant positive cases. Finding the right mix between recall and precision is crucial. Consider the total influence on the problem you’re tackling because a high-precision model may miss crucial occurrences.

Recall measures how many of the actual positive instances were correctly predicted. Focusing solely on recall and ignoring precision can lead to a large number of false positives. This can happen when a model predicts most instances as positive, leading to high recall but low precision. Similar to the previous point, ensure you consider both precision and recall. The high recall might lead to excessive false positives, affecting the model’s practicality.

3- The harmonic mean of recall and precision, which is the F1 score, offers a fair comparison of the two. The most frequent mistake is to think that the F1 score is only an average of recall and precision. The F1 score is sensitive to imbalances between precision and recall since it favors the lesser of the two values. The F1 score is the harmonic mean of recall and precision, and it favors the lower number more than the higher one. It helps you strike a balance between recall and precision. It’s not a straightforward average, so be careful.