Early Prediction of Complex Business Processes Using Association Rule Based Mining

Abstract

Complex business processes are challenging and hard to analyse. The objective here is to enhance delivery of processes in terms of improving quality of service and customer satisfaction. Therefore, an automated process prediction system is desirable to monitor and evaluate complex business processes and forecast process outcome during execution time. The analysis of such processes would help domain experts to make in-time decisions to improve the process. The in-time response greatly effects the quality of service and customer satisfaction. Therefore, in this paper, the early process prediction framework using Classification Based on Association rules (CBA) has been proposed to predict outcomes for such incomplete processes. The essential part of the proposed system is to extract association rules from the process data up to a certain point in time (i.e. the cut-off time) at which the prediction needs to be made; in an live process this would usually be the current time. The CBA algorithm generates rules with user specified support and confidence which are then utilised for early process prediction. The experimental results based on real business process data are presented for on-time and delayed processes. The proposed early process prediction system is evaluated using different metrics such as accuracy, precision, recall and the F-measure. Moreover, the proposed system is also compared with our prior published work in terms of accuracy, recall and F-measure. The analysis shows that the performance of proposed system outperforms schemes in the literature.