With the development of information technology,information systems are widely used in a variety of organizations.People,machine and software will leave tracks in the information system when executing activities,these tracks were recorded in the so-called event logs.Event log mining(also known as process mining)aims at mining business process models from these logs automatically.After comparing with existing models to diagnose and detect deviations,the mining models present new insights to help enterprises get promoted.Event log mining is a relatively young field.On the one hand,it is between machine learning and data mining.On the other hand,it links process modeling with process analysis.Currently,event log mining has attracted the attention of researchers from business process management and software vendors.Scholars have put forward a variety of mining algorithms,software vendors also launched corresponding function in their products.Comparing with the traditional process modeling,event log mining builds models objectively by mining the historical data and works as X-rays to diagnose problems in the process to find the solutions.Therefore,event log mining has a certain theoretical and practical significance.Firstly,this paper introduces the background and significance of the event log mining and summarizes the relevant research achievements at home and abroad,then points out the advantages and disadvantages of current event log mining algorithms.The concepts of event log mining includes the definition of process mining and event logs,Petri net and causal matrix as well as their relationship.The established models and the implementation of previous research have laid a good foundation for our research on event log mining.Secondly,this paper improves the traditional genetic algorithm on event log mining by designing the fitness function and proposing the heuristic rules to initialize population.We add heuristic rules to cross point selection in order to improve the fitness value of genetic individuals after crossover operation.Benefiting from the decay mutation rate,the algorithm can reduce the diversity of the population appropriately in later stage and thus accelerating the algorithm converge to the optimal solution.Moreover,we come up with a hybrid mining algorithm based on previous improved genetic mining algorithm.Using simulated annealing algorithm and bees breeding theory to help genetic algorithm to retain the best individual genes while avioding the inefficiency and premature convergence of original genetic algorithm.And thus improving the performance of hybrid genetic mining algorithm.Eventually,we use a certain number of event logs to test the minng algorithms including improved genetic mining algorithm and hybrid genetic mining algorithm on ProM platform.Through the experiments on ProM platform,we obtained the testing results which containing running time of algorithm and five quality criteria in the process mining field.By analyzing the data of five main quality dimensions,we found that the improved algorithm this paper proposed perform well than the original one in mining models from event logs.Furthermore,the hybrid one perform better than the improved one.Expecially,two kinds of mining algorithms this paper proposed outperform than original genetic mining algorithm in mining more complex log obviously. |