Ways to Evaluate Classification Algorithms

With video explanation | Data Series | Episode 10.3

There are various classification algorithms such as:

• Support Vector Machines (Episode 9.1)
• Logistic Regression (Episode 7.1)
• Decision Trees (Episode 11.1)

But how do we evaluate these algorithms’ performance?

In this episode we look at the following evaluation metrics:

1. Accuracy, Precision, Recall (True Positive Rate), False Positive Rate, Specificity, Sensitivity, F1 score.
2. AUROC Score (Area Under the Receiver Operator Characteristic)

Let’s suppose we had some data and put the data in a model that predicts either positive or negative:

How well did this model perform? There are many different ways we can look at this. First, let us look at our model’s predictions in a matrix form. This is known as a confusion matrix: