| Abstract | This thesis addresses contextual and ethical issues in the predictive process monitoring framework and several related issues. Regarding contextual issues, even though the importance of case, process, social and external contextual factors in the predictive business process monitoring framework has been acknowledged, few studies have incorporated these into the framework or measured their impact. Regarding ethical issues, we examine how human agents make decisions with the assistance of process monitoring tools and provide recommendation to facilitate the design of tools which enables a user to recognise the presence of algorithmic discrimination in the predictions provided.
First, a systematic literature review is undertaken to identify existing studies which adopt a clustering-based remaining-time predictive process monitoring approach, and a comparative analysis is performed to compare and benchmark the output of the identified studies using 5 real-life event logs. This curates the studies which have adopted this important family of predictive process monitoring approaches but also facilitates comparison as the various studies utilised different datasets, parameters, and evaluation measures.
Subsequently, the next two chapter investigate the impact of social and spatial contextual factors in the predictive process monitoring framework. Social factors encompass the way humans and automated agents interact within a particular organisation to execute process-related activities. The impact of social contextual features in the predictive process monitoring framework is investigated utilising a survival analysis approach. The proposed approach is benchmarked against existing approaches using five real-life event logs and outperforms these approaches. Spatial context (a type of external context) is also shown to improve the predictive power of business process monitoring models.
The penultimate chapter examines the nature of the relationship between workload (a process contextual factor) and stress (a social contextual factor) by utilising a simulation-based approach to investigate the diffusion of workload-induced stress in the workplace.
In conclusion, the thesis examines how users utilise predictive process monitoring (and AI) tools to make decisions. Whilst these tools have delivered real benefits in terms of improved service quality and reduction in processing time, among others, they have also raised issues which have real-world ethical implications such as recommending different credit outcomes for individuals who have an identical financial profile but different characteristics (e.g., gender, race). This chapter amalgamates the literature in the fields of ethical decision making and explainable AI and proposes, but does not attempt to validate empirically, propositions and belief statements based on the synthesis of the existing literature, observation, logic, and empirical analogy. |
|---|