Confusion Matrix

Now let's start with performance metrics for a classification problem. In this lesson, we focus on describing metrics for evaluating binary classification problems, for example predicting whether patients would develop heart failure or not. These metrics can be generalized to the setting with multiple classes as well. But in this lesson, we'll focus on the binary setting. There are many evaluation metrics for classification model. Next, we'll illustrate how they're all connected. To depending on the prediction and the Ground Truth value, we have four different possible outcomes. True Positive, False Positive, False Negative and True Negative. Next, we explain their definition and relationship to each other. So, the output of a binary predictive model, can be either Positive or Negative. Here, you can see a zoom in version of this. To differentiate with the Ground Truth value, we call this Prediction Outcome Positive, and Prediction Outcome Negative. Similarly, we have the Ground Truth value can be either Positive and Negative. Again, to differentiate with the prediction value, we call this Condition Positive and Condition Negative. If the Prediction Outcome is positive, and the Ground Truth's value is also positive, then we call this, True Positive. However, if the Prediction Outcome is positive, but the Ground Truth condition is negative, then we call this False Positive, or Type I error. Conversely, if the Prediction Outcome is negative, and the Ground Truth Condition is positive, we call this False Negative, or Type II error. Finally, a True Negative occur, when Prediction Outcome and Ground Truth Condition are both Negative. In fact, this 2x2 matrix is called a Confusion Matrix, or a Contingency table.