These graphs show the aggregated classification over three different runs, where for every category (FN, TP, TN, FP) the minimal number over the different runs is taken. For example, if 10 traces are TP in the first run, 8 in the second, and 9 in the third, then the value 8 is used as aggregated value. As a result, these graphs can show discrepancies between the classifications in different runs.
For the Hybrid ILP Miner, there is a discrepancy (note that the bar for pdc_2106_5.xes do not reach 20). The Hybrid ILP Miner discovered different nets from this log over the three runs.