First off, why do we even need these fancy numbers to measure how well our models are doing? Well, because sometimes they don’t perform as expected and we want to know where we went wrong. And let’s be real here, who doesn’t love a good number game?!

So, what metrics do we use for classification tasks? There are many options out there, but the most popular ones include accuracy, precision, recall, and F1 score. Let’s break them down one by one:

Accuracy is the simplest metric to understand it tells us how often our model correctly classifies a data point as belonging to either class A or B (or C, D, E…). It’s calculated using this formula:

(true positives + true negatives) / total number of samples

Precision is the ratio of true positive predictions to all positive predictions. In other words, it measures how often our model correctly identifies a data point as belonging to class A when it actually does belong to that class (i.e., no false positives). It’s calculated using this formula:

true positives / (true positives + false positives)

Recall is the ratio of true positive predictions to all actual positive samples in our dataset. In other words, it measures how often our model correctly identifies a data point as belonging to class A when it actually does belong to that class (i.e., no missed detections). It’s calculated using this formula:

true positives / (true positives + false negatives)

F1 score is the harmonic mean of precision and recall, which takes into account both false positives and false negatives. It’s calculated using this formula:

2 * ((precision * recall) / (precision + recall))

Now that we know what these metrics are, how to interpret them. A high accuracy score is always a good thing it means our model is doing well overall. However, if the dataset has an imbalanced distribution of classes (i.e., one class is much more common than the other), then we might want to focus on precision and recall instead.

For example, let’s say we have a dataset where 95% of samples belong to class A and only 5% belong to class B. If our model has an accuracy score of 98%, that sounds pretty good but what if it’s because the model is always predicting class A? In this case, precision would be much more important than accuracy since we don’t want false positives (i.e., incorrectly labeling a sample as belonging to class B when it actually belongs to class A).

Remember, always keep an eye out for those ***** false positives and false negatives .