For the given dataset, if students took improvement exams (i.e, rewrote them) and hence wrote fewer than five subjects, the average of those subjects alone were taken. In all other cases like absenteeism in an exam for which the student has been registered for, a score is 0 is taken in this case. This means that it is not possible to distinguish from the graph - though it is from the dataset - whether a student actually scored zero or did not show up for the exam itself.