Data Science · Maastricht
Knoors Data Science
Data science for domains where accuracy,
sensitivity and rigour matter.
Data Science · Maastricht
Data science for domains where accuracy,
sensitivity and rigour matter.
About
Knoors Data Science is a data science practice based in Maastricht, Limburg. The focus is on organisations that need to get more out of their data — building models from scratch, integrating AI into existing workflows, or handling sensitive data in a way that is technically sound and ethically defensible.
Past projects span national cancer registries, regional government and early-stage companies, with work published in peer-reviewed journals and presented at international conferences. Daan also contributes through Dev For Good, a collective working on technical projects with a social or environmental purpose.
Services
Predictive models for classification, regression, survival analysis and anomaly detection — from problem definition through to deployment.
Decision-support tools, document processing pipelines, automated reporting and LLM-based applications tailored to your domain.
Risk scoring, outcome prediction and diagnostic support for clinical and public health questions in regulated, data-sensitive settings.
Technical solutions for sensitive data — de-identification, differential privacy, federated learning, synthetic data and secure record linkage.
Privacy Engineering
Many organisations hold data they cannot use — because it contains personal information, or because sharing it creates legal and ethical risk. The work is building technical solutions that unlock that data without compromising privacy.
Removing or replacing identifiers to reduce re-identification risk while keeping the data analytically useful.
Adding calibrated noise to outputs so that individual records cannot be inferred from results.
Producing artificial datasets that mirror the statistical structure of real data, suitable for sharing and testing.
Training models across multiple data holders without any party sharing raw data.
Joint computation over private inputs, where no party learns anything beyond the agreed output.
Privacy-preserving merging of databases using probabilistic matching on encrypted data.
Selected work
A classifier trained on the Netherlands Cancer Registry to predict metachronous metastases after primary breast cancer diagnosis. Required probabilistic record matching across multiple hospital databases. Average precision of 0.95, presented at the San Antonio Breast Cancer Symposium.
LSTM and GRU networks trained on longitudinal screening histories to predict cervical cancer risk, validated across Norwegian and Estonian screening programmes. Published in the International Journal of Medical Informatics.
A synthetic version of the Netherlands Cancer Registry that preserves the statistical structure of the original without containing real patient records — one of the first high-dimensional medical synthetic datasets released publicly for research.
Joint analyses across cancer registries in the Netherlands and the Nordic countries — institutions training shared models without centralising data. Combined federated learning, secure multi-party computation and differential privacy. Published in Frontiers in Oncology.
Clients
Contact
Available for project-based and longer-term engagements across the Netherlands and beyond. Got a concrete problem, or just want to explore what's possible? Feel free to reach out.
info@knoorsdatascience.nlBased in Maastricht, Limburg — working nationally and internationally.