When you start learning about machine learning, it's really important to understand how we measure how well our models work. One way to do this is by looking at accuracy.

Accuracy shows you the percentage of correct guesses made by your model. You can think of it like this:

$Accuracy = \frac{TP + TN}{TP + TN + FP + FN}$

Let’s break that down:

• TP stands for true positives (correctly guessed items).
• TN means true negatives (items that were correctly identified as not belonging to a certain class).
• FP is false positives (items that were guessed wrong, saying they belong to a class when they don’t).
• FN is false negatives (items that belong to a class but were missed).

This formula gives us a simple way to see how the model is doing overall.

But, there’s a problem if we only look at accuracy. This can trick us, especially when our data isn’t balanced, meaning one class is much bigger than the others.

For example, if 90% of your data is from one group, a model that always guesses that group could seem accurate 90% of the time, even though it doesn’t really help us at all!

That’s why we also use other measurements like precision, recall, and the F1 score.

• Precision shows how many of the selected items were actually correct.
• Recall tells us how many of the correct items were found.
• F1 Score is a way to take both precision and recall and find a good balance between them.

Finally, there's ROC-AUC. This is a nice way to visualize how our model is doing by showing the balance between true positives and false positives, which helps us understand model performance better.

From my experience, using these different measurements together gives us a clearer picture of how well our model is working. Plus, it helps us improve our models for the best results!