An application of quantum machine learning for inter-case predictive process monitoring

Today, complex business decisions are based on big data. The omnipresent digital infrastructure allows to log events and extract visual representations from complex business processes. In that manner, it is possible to track business cases and convoluted interactions in real-time. Furthermore, under the umbrella of Predictive Process Monitoring (PPM), a number of machine learning techniques emerged that aims to forecast future behaviour of a business case such as next activities or remaining time. Processes occur everywhere and applications for PPM can be found in domains ranging from logistics to energy management to hospitals. However, often, the outcome of a single process instance does not solely depend on its history rather than the state of other instances that are executed in the same time. Likewise, feature vectors for accurate recommendations grow arbitrarily for a huge amount of process instances.

This poster presents a novel technique to tackle the increasing complexity twofold by (a) investigating ways to encode inter-case dependencies and (b) using advanced techniques to deal with higher-dimensional feature spaces. Here, quantum algorithms are expected to provide an advantage over classical methods when calculating the inner product in higher dimensional Hilbert Spaces. In the evaluation, classical and quantum computational methods are benchmarked on several combinations of inter-case data encodings and real-world event log datasets. Experiments show that the familiy of quantum kernel-based algorithms outperforms the classical ones in terms of accuracy with scaling amount of features which gives hope to achieve quantum advantage in the near future.