What are some common evaluation metrics for classification models, and how would you choose which one to use?

In my final year project, I developed a classification model to predict student performance based on various features like attendance, assignments, and exam scores. To evaluate my model, I used accuracy as the primary metric because my dataset had a balanced distribution of pass and fail students. However, I also calculated precision and recall to ensure that my model wasn't just predicting the majority class. This experience taught me the importance of not only looking at one metric but considering multiple aspects to get a fuller picture of the model's performance.

While volunteering for a non-profit organization, I assisted in a campaign to classify potential donors based on their likelihood to contribute. We utilized precision and recall to evaluate our model since it was crucial to identify as many willing donors as possible without falsely labeling too many uninterested ones. This experience highlighted how different metrics can serve specific purposes, and I learned to choose evaluation metrics based on the goals of the project rather than a one-size-fits-all approach.

During my internship at a tech startup, I worked on a customer churn prediction model. We prioritized the F1 score as our evaluation metric because it provided a balance between precision and recall. Given our goal was to minimize customer loss, focusing solely on accuracy would have been misleading, as our dataset was imbalanced. This experience helped me understand that in a real-world setting, the choice of evaluation metric could significantly affect business decisions.