Conformance checking and diagnosis in process mining

摘要: In the last decades, capability of information systems to generate and record overwhelming amounts event data has experimented an exponential growth in several domains, particular industrial scenarios. Devices connected internet (internet things), social interaction, mobile computing, cloud computing provide new sources this trend will continue next decades. The omnipresence large stored logs is important enabler for process mining, a novel discipline addressing challenges related business management, modeling, intelligence. Process mining techniques can be used discover, analyze improve real processes, by extracting models from observed behavior. these represent reality determines quality results obtained them, conditioning its usefulness. Conformance checking aim thesis, where modeled behavior are analyzed determine if model defines faithful representation log. Most efforts conformance have focused on measuring ensuring that capture all log, i.e., fitness. Other properties, such as precise (not including unnecessary behavior) been disregarded. first part thesis focuses analyzing precision dimension conformance, describing precisely preferred overly general models. includes technique based detecting escaping arcs, points deviates one reflected detected arcs determine, terms metric, between log model, locate possible actuation order achieve more model. also presents confidence interval provided multi-factor measure assess severity imprecisions. Checking time consuming real-life scenarios, understanding reasons behind mismatches effort-demanding task. second changes focus fitness dimension, proposes use decomposed aid diagnosing proposed approach decomposing into single entry exit components. resulting fragments subprocesses within main with simple interface rest Fitness per component provides well-localized information, aiding diagnosis causes problems. Moreover, relations components exploded capabilities analysis, identifying areas high degree mismatches, or providing hierarchy zoom-in zoom-out analysis. Finally, two applications approach. First, theory extended incorporate manner. Second, real-time event-based framework presented monitoring