How DataCore Analytics Keeps Data Confidential
12 May 2026
Machine learning only earns its place when it beats a simple baseline. We test that first, and we tell you when it does not.
A great deal of published health machine learning reports impressive accuracy on data that leaked between training and test sets, or on cohorts that look nothing like the population the model would be used in. Models trained overwhelmingly on European and North American data are a particular problem for African clinical deployment.
We start every project with the simplest sensible baseline — logistic regression, an existing clinical score — and only move to complex models if they measurably beat it under honest validation.
Each project uses the subset of these that your research question requires.
Before modelling: is there signal, is the sample size adequate for the number of features, and how does a simple model perform? Sometimes the honest answer is that the data cannot support the model you want.
Principled feature construction and selection with the selection step inside the cross-validation loop, so performance estimates are not inflated by leakage.
Regularised regression, gradient boosting, random forests and neural networks where the data justifies them, with nested cross-validation and hyperparameter tuning done properly.
Held-out and external validation, calibration assessment, decision curve analysis, and subgroup performance so you know where the model fails and for whom.
SHAP values, feature importance and partial dependence, so a clinician or reviewer can see what the model is actually using.
Packaged inference code, a container, and documentation for teams who want to run the model in their own environment.
Typical turnaround: 4–10 weeks depending on data complexity and validation requirements
Indicative price: Project-quoted; feasibility assessment from $499
Reduced rates are available for students and researchers at African public institutions. Every project is quoted in writing before work begins.
Services are provided for research purposes only. They are not intended for clinical diagnosis, treatment decisions or individual health assessment. See how it works, data submission guidelines and what you receive.
Bioinformatics, biostatistics and data science support for research teams across Africa.
12 May 2026
28 April 2026