---
title: "Essays on Predictive and Prescriptive Process Monitoring (PhD Thesis)"
type: source
tags: [ppm, prescriptive-process-monitoring, remaining-time, simulation, concept-drift, customer-service, loyalty, queuing, thesis, meta-ingest]
authors: [Riess, Mike]
year: 2023
venue: "PhD thesis, Norwegian University of Life Sciences (NMBU), School of Economics and Business, Thesis 2022:21, ISBN 978-82-575-1896-7"
kind: book
raw_path: "raw/Riess/PhD Thesis Mike Riess (2023).pdf"
sources: []
key_claims:
  - "Four-essay thesis on proactive decision support in business processes: earliness of remaining-time prediction (Paper I), synthetic simulation for PPM evaluation (Paper II), predictive queue prioritisation for customer loyalty (Paper III), and metaheuristics for concept-drift adaptation (Paper IV)."
  - "Paper I introduces Temporal Consistency (TC) as a new performance dimension for remaining-time prediction alongside accuracy and earliness, and proposes three L1 losses with temporal decay (exponential MAEEtD, power MAEPtD, moderate MAEMtD)."
  - "Paper II argues that ecological-validity-first BPS frameworks cannot answer data-generating-process hypotheses; proposes a parametric framework (SynBPS) where process memory, entropy, activity durations and stability are user-controlled."
  - "Paper III (with Scholderer) proposes predicted-loyalty queue prioritisation via NPS-conditional cycle-time prediction in a customer service process of a European telecom; evaluated via agent-based simulation calibrated on historical data."
  - "Paper IV surveys metaheuristics for concept-drift adaptation; finds population-based methods (GA, PSO) dominate and that evaluation rigour (drift characterisation, class-balance-aware metrics) is often lacking."
  - "Methodological through-line: evaluation validity (external and internal) is treated as the central problem of PPM research, motivating both the simulation framework (Paper II) and the drift-adaptation survey (Paper IV)."
  - "Supervisor: Joachim Scholderer. Defended 2022; published 2023."
created: 2026-04-20
updated: 2026-04-20
---

# Essays on Predictive and Prescriptive Process Monitoring (PhD Thesis, 2023)

Mike Riess's PhD thesis at [[entities/nmbu|NMBU]] School of Economics and Business, defended 2022, published 2023. The thesis bundles four essays (three single-authored, one co-authored with supervisor Joachim Scholderer) under the umbrella of **proactive decision support in business processes** — predictive process monitoring and prescriptive process monitoring.

## Summary
The thesis frames PPM/PrPM as a response to the insufficiency of reactive tools (SPC, Lean/Six Sigma, workforce management) once business processes are heavily digitised through process-aware information systems ([[concepts/process-aware-information-system|PaIS]]). Value creation is measured along the *devil's quadrangle* dimensions (time, cost, quality, flexibility; [[concepts/devils-quadrangle]]). Machine learning on [[sources/2023-berti-ocel-2-specification|event-log]] data is the principal vehicle.

Four research objectives:
1. Understand how temporal weighting of loss functions affects earliness performance of remaining-time models.
2. Propose a simulation framework enabling researchers to specify and generate synthetic event logs for controlled-variable PPM evaluation.
3. Understand how customer loyalty can be influenced by predictive queue prioritisation in customer service.
4. Review how prior literature has handled automated adaptation of ML models under concept drift.

## Essay index (with mapping to separately ingested papers)

| # | Title | Status in thesis | Stand-alone ingested source |
|---|---|---|---|
| I | Remaining cycle time prediction: Temporal loss functions and prediction consistency | Submitted to Nordic Machine Intelligence (published 2023) | [[sources/2023-riess-temporal-loss-remaining-cycle-time]] |
| II | A parametric simulation framework for the generation of event-log data | Submitted to SIMULATION (published 2024 as SynBPS) | [[sources/2024-riess-synbps-simulation-framework]] |
| III | Customer-service queuing based on predicted loyalty outcomes (with J. Scholderer) | Submitted to Decision Support Systems | *Not in this ingest batch — referenced-only* |
| IV | Automating model management: A survey on metaheuristics for concept-drift adaptation | Revised version of JDIM 2022 paper | [[sources/2022-riess-metaheuristics-concept-drift-survey]] |

Paper III is referenced through the thesis summary only; the underlying manuscript is not present in `raw/` and is therefore filed as *referenced-not-ingested*.

## Structural themes
- **Earliness and temporal consistency** (Paper I): predictions from open cases must be accurate early and must follow the natural monotonic decrease of remaining time. Three L1 variants (exponential, power, moderate temporal decay) are tested across four public event logs (Sepsis, Helpdesk, Traffic fines, Hospital billing).
- **Controlled synthetic evaluation** (Paper II): existing calibrated BPS tools optimise for ecological validity and cannot isolate data-generating-process factors; hence a purely parametric Markov-chain-based framework is proposed.
- **Prescriptive queue prioritisation** (Paper III): predicted cycle time → predicted NPS → dynamic priority; evaluated against FCFS, SRTF, LRTF via [[methods/process-simulation|agent-based simulation]]. Predicted-loyalty approach performs similar to LRTF (both depend on cycle-time prediction) and collapses to FCFS once a 60-hour service level is enforced.
- **Drift adaptation** (Paper IV): population-based metaheuristics dominate; drift characterisation in evaluation is often absent.

## Methodological commitments
- Preference for **publicly available event logs** as benchmarks (Sepsis, Helpdesk, BPIC Traffic fines, Hospital billing) — with explicit criticism of this practice in Paper II.
- LSTM as the PPM workhorse, following [[sources/2017-tax-lstm-process-prediction|Tax et al. 2017]] and [[sources/2017-navarin-lstm-data-aware-remaining-time|Navarin et al. 2017]].
- Simulation as a methodology both for evaluation (Paper II) and as a testbed for prescriptive hypotheses (Paper III).
- Literature-review methodology in Paper IV based on Google Scholar with explicit inclusion/exclusion criteria (snapshot-in-time caveat acknowledged).

## Connections
- [[entities/mike-riess]] — author hub.
- [[entities/nmbu]] — host institution; supervisor [[entities/joachim-scholderer|Joachim Scholderer]] *(unverified entity page; may be created)*.
- [[concepts/predictive-process-monitoring]], [[concepts/remaining-time-prediction]], [[concepts/concept-drift]], [[concepts/business-process-simulation]].
- [[syntheses/riess-research-arc]] — synthesis across all seven Riess sources.
- Cited works in this wiki: [[sources/2017-tax-lstm-process-prediction]], [[sources/2017-navarin-lstm-data-aware-remaining-time]], [[sources/2016-teinemaa-structured-unstructured-ppm]], [[sources/2019-verenich-survey-ppm]], [[sources/2020-rama-maneiro-deep-learning-ppm-review]], [[sources/2018-difrancescomarino-genetic-hpo-ppm]], [[sources/2018-dumas-fundamentals-of-bpm]], [[sources/2011-vanderaalst-process-mining-book]].