Confusion matrix is a square matrix used to assess the accuracy of classification methods. The rows are labelled by the class given by the classifier and the columns are labelled by the actual class. Each cell is the probability (or sometimes percentage or freqeuncy) with with which an example that is classified as a partcular class is actually of another. The diagonal represents correct classifications and the off-diagonal elements errors. The numbers on a row shoudl add up to 1 (for probabilities) or 100% (for percentages). In order to turn this information into actual accuracy fifgures, one needs to know about the base rate for each class. For example, the diagonal entries could be largely near 100% (highly accureate) for all but one case, but that case might be the most common in practice.
Used on page 466
Table 19.2 Confusion matrix for activity recognition