Predictive process mining meets computer vision

Abstract

Nowadays predictive process mining is playing a fundamental role in the business scenario as it is emerging as an effective means to monitor the execution of any business running process. In particular, knowing in advance the next activity of a running process instance may foster an optimal management of resources and promptly trigger remedial operations to be carried out. The problem of next activity prediction has been already tackled in the literature by formulating several machine learning and process mining approaches. In particular, the successful milestones achieved in computer vision by deep artificial neural networks have recently inspired the application of such architectures in several fields. The original contribution of this work consists of paving the way for relating computer vision to process mining via deep neural networks. To this aim, the paper pioneers the use of an RGB encoding of process instances useful to train a 2-D Convolutional Neural Network based on Inception block. The empirical study proves the effectiveness of the proposed approach for next-activity prediction on different real-world event logs.