Fig. 4

Execution of the algorithm, showing convergence in iteration 7. Each line represents the difference in distribution for a particular sub-hierarchy of a balanced module at a given iteration, when compared to SNOMED CT. For example, in the balanced module at iteration 1, Body Structure is proportionally 11.26 % bigger than in SNOMED CT (it is over-represented), while Clinical Finding is 6.86 % (it is under-represented). The dashed line represents the residual sum of squares of all sub-hierarchies, 0 meaning that the sub-hierarchies in the balanced module are distributed in exactly the same way as in SNOMED CT (perfectly balanced module)