Confusion matrix, using real data from the cross validation dataset, for the three-class model. Labels as in Fig. 4. The binned and phase-folded input data from a single randomly selected object are shown in each subplot, with the light curve in blue and the centroid curve in orange. The black number is the total number of objects classified in this subset, while the red number shows recall to two significant figures, i.e. the proportion of each class that is estimatedto be a member of this class (hence horizontal rows always sum to 100% within rounding errors). Objects on the diagonal are correctly classified and coloured green, while those outside are mis-classified and are coloured red. The strength of the background colours is proportional to the percentage (i.e. recall) in each box.