Abstract
Alignments provide sophisticated diagnostics that pinpoint deviations in a trace with respect to a process model. One crucial aspect is to consider, in the alignment task, not only the control flow perspective but also other sources of information available in event logs like data payloads. However, the combination of these dimensions makes the problem of multi-perspective trace alignment highly challenging since the number of traces accepted by the model is typically infinite. In this paper, we address this problem by proposing an approximate approach to alignment computation: instead of computing the optimal alignments based on the complete knowledge about a process trace available in the log, we perform approximate alignments based on lossy trace encodings that only consider certain information about the trace. The advantage of this approach is twofold. First, the trace alignment task is much faster. Second, the analyst can choose what type of information is relevant for computing the alignments by selecting the encodings that represent a trace based on that information. Our experiments show that the approximate approach is faster than the optimal one and, for encodings sufficiently rich, able to provide accurate results.