Johannes Mäkelburg

Research profile

Research summary

Johannes Mäkelburg is an early-stage researcher at the Technical University of Munich working at the intersection of applied machine learning and information systems engineering. His record splits evenly (0.5/0.5) between machine learning for cyber-physical systems and business process mining / predictive process monitoring. In the first strand he is first author of "Estimating Inconsistency Response Surfaces under Uncertainty in Cyber-Physical System Development" (2026, arXiv), which learns surrogate models of CPS inconsistency response surfaces reported as roughly 935× faster than Monte Carlo sampling, drawing on zonotopes and multi-fidelity Monte Carlo; Evolve rates it high reproducibility readiness. In the second strand he contributed to GNN4PPM (2026, arXiv), applying relational graph convolutional networks over RDF knowledge graphs to multi-target next-event prediction evaluated on six BPIC logs. His OpenAlex topic profile is broad: Semantic Web and Ontologies (5 works), Data Quality and Management (4), Systems Engineering Methodologies and Applications (4), Business Process Modeling and Analysis (3), and Probabilistic and Robust Engineering Design (3). Output is rising steeply — 1 work in 2024, 5 in 2025, 11 in 2026, 17 over the last three years — while the collaboration network remains small (4 distinct coauthors). Citation-based impact cannot yet be assessed: the snapshot reports an h-index of 3 and an i10-index of 0. Evolve has reproducibility data for 2 papers, with a mean readiness of 78.

Research keywords

Highlights

• Surrogate models ~935× faster than Monte Carlo for CPS inconsistency analysis

  As first author, he helped build a surrogate that learns inconsistency response surfaces under uncertainty in cyber-physical system development, reported at roughly 935× the speed of Monte Carlo sampling, using zonotopes and multi-fidelity Monte Carlo. Evolve calls it well-benchmarked with high reproducibility readiness, though only pairwise low-dimensional settings were tested.

• R-GCN over knowledge graphs for multi-target process monitoring

  GNN4PPM frames multi-target next-event prediction in predictive process monitoring as learning over RDF graphs with relational graph convolutional networks, achieving best or competitive next-activity prediction (NAP) results across six BPIC logs. Evolve notes the evidence is weakened by a validation-only evaluation, and rates reproducibility readiness medium.

• Steep output ramp with solid reproducibility on reviewed work

  Indexed output grew from 1 work (2024) to 5 (2025) to 11 (2026) — 17 works in three years — while the two Evolve-reviewed papers average a reproducibility readiness of 78, with one rated high. A productive early-career trajectory whose main gaps are impact evidence and a still-small coauthor network.

Timeline

1. 2024First indexed work appears in the OpenAlex record (1 work in the counts-by-year series).
2. 2025Output expands to 5 indexed works as the record builds across both research strands.
3. 2026Most active year to date: 11 indexed works, bringing the last-three-year total to 17.
4. 2026Two arXiv preprints enter the Evolve catalog: the first-authored CPS inconsistency surrogate paper (high readiness) and GNN4PPM on relational graph networks for process monitoring (middle authorship, medium readiness).

Papers in the Evolve catalog

• 2026-09-13

  GNN4PPM: Multi-Target Predictive Process Monitoring with Relational Graph Convolutional Networks

• 2026-09-10first author

  Estimating Inconsistency Response Surfaces under Uncertainty in Cyber-Physical System Development

Frequent co-authors