Process Mining Algorithms Performance According to New Bayes Conformance Function

Abstract

Process mining includes discovery of processes, analyzing them, modeling and improvement studies and thus enables efficient and effective use of process data in this way. It became essential for big companies to employ process mining techniques in competitive business environment. This study is consisted of two stages. In the first stage, a system displaying and analyzing various properties of synthetic and real data which may assist in modeling and developing processes. Under the scope of this study, a system employable with various process models and meaningful information from such models are produced. Thanks to this system, more productive and efficient processes can be developed. In the second stage, process discovery is performed by using algorithms employed in this field with synthetic and real event logs. Performances of algorithms are compared by employing conformance function, accuracy rate and time criteria by displaying discovered processes with petri networks. In our proposed model, we created the new conformance function with naive bayes. As a result, more successful results are obtained with heuristic algorithms in terms of likelihood function and accuracy rate than alpha algorithm. However, alpha algorithm is more advantageous than heuristic algorithm in terms of time. Since genetic operators (selection, crossover, mutation) are used, genetic algorithm provided better successful results than alpha and heuristic algorithms.