Medical diagnosis classifier

What this project does

This small teaching project trains an interpretable baseline classifier for a binary medical diagnosis task using a licensed, de-identified teaching dataset included with scikit-learn. The goal is educational: demonstrate careful data handling (explicit missing-data treatment and patient-level separation), train a simple logistic regression baseline, and report clinically-relevant metrics (sensitivity, specificity) together with calibration and uncertainty estimates. The code is intentionally sandboxed and does not provide patient-specific diagnosis or treatment advice.

Data

This project uses the scikit-learn breast cancer dataset (diagnostic Wisconsin). That dataset provides de-identified, per-patient numeric features and a binary target (malignant/benign). No external data are used and no patient identifiers are exposed.

How to build it

1. Create a reproducible Python environment with the required packages (see requirements).
2. Open the project file project.py and review the preprocessing: missing-value imputation, scaling, and the logistic regression baseline.
3. Run the pipeline locally to produce held-out metrics and subgroup checks.
4. Inspect results and the README notes about limitations before drawing any clinical conclusions.

How to run

Run the project module directly. It prints a JSON summary of metrics to stdout.

How to test

Run pytest in the project directory. Tests check that the run() function returns the expected keys, that numeric metrics are within valid ranges, that a conservative performance floor is met on the held-out split, and that results are deterministic for a fixed seed.

Results

Measured by running python project.py (seed 42) on 2026-10-23; the tests in test_project.py assert the floor.

Metric	Value
accuracy	0.9650
sensitivity	0.9245
specificity	0.9889
brier	0.0215
ece	0.0323
subgroup_sensitivity_diff	-0.0816
model	LogisticRegression

Also returned (see python project.py output): sensitivity_ci (2 items), specificity_ci (2 items).

Limitations

This exercise is educational. Even though the dataset is a real, de-identified teaching dataset, retrospective performance on this dataset does not establish diagnostic safety or clinical benefit. The model is a simple logistic regression baseline and has had no clinical validation, prospective testing, nor assessment of harms. The pipeline also uses a single held-out test split; a full evaluation would include external validation, robust calibration assessment in new groups, and clinician review.

Next steps

Suggested next steps for learning: (1) add nested cross-validation and repeated runs to better estimate uncertainty; (2) evaluate external datasets or temporal splits; (3) include clinician adjudication for any ambiguous labels; (4) explore simple explainability methods (feature coefficients, SHAP-lite analyses) and assess fairness across subgroups before considering any clinical translation.

---

Full case study, build steps and evaluation checks: Medical diagnosis classifier on networkershome.com

Built by Networkers Home · Learn the skills in the AI Engineer Course · Part of 100 AI projects with code