Estimating Remaining Time of Business Processes based on Traces Structural Analysis

Prediction of the remaining times of a business process execution is one of the main challenges of process mining, being critical for many organisations. In this work we present a new approach based on the structural analysis of partial traces, each one is associated with a state of the transition system that describes the log behaviour. From this analysis, a set of parameters or attributes that characterise each partial execution of the process is obtained. The hypothesis behind it is that the remaining time to finish the process execution depends on these structural features. Once this set of attributes is obtained for each transition system state, using of the linear regression technique is utilized to predict the remaining time of the business process for each state. Experiments with eight well-known real-life datasets show that our approach outperforms the baseline model in the three metrics considered. We also analysed the performance of our approach when combined with two attribute selection methods, in order to demonstrate the estimation accuracy in regard to the time consumed in the calculation of the remaining time estimation. Experimental results indicate that the Forward Best-First method provides a good balance between estimation accuracy and computation time consumed.