The aim of the confusion matrix is to give you clues for improving your classifier and your data quality. In the confusion matrix, the categories are analyzed in greater detail. It also gives you more transparency with regard to the results of Precision, Recall and F1-Mass.
The confusion matrix is formed from the two axes «Actual category» and «Predicted category». The precision value can be calculated from the predicted categories (= diagonal element/sum of the column), and the recall value from the actual categories (= diagonal element/sum of the row).
The blue diagonal depicts the amount of the correctly classified examples per category. In all the other table cells, the incorrect examples are classified. The other differently colored fields indicate a particularly high number of examples that could not be classified correctly.
The confusion matrix offers two possibilities for interaction:
- The mouse-over gives you an overview of which category the examples in a field belong to and what the category-prediction for these examples is.
- At the same time, the inverse field is highlighted. These are the same categories in reversed order. By clicking on a cell, the examples it contains will appear beneath the confusion matrix.