Rare Disease Detection: SMOTE Class-Imbalance Correction (2D)

Diagnostic machine-learning models for rare conditions face a structural problem before any algorithm is chosen: the disease is, by definition, rare in the training data. A classifier that sees 160 healthy patients for every 14 real cases can reach 92% accuracy by never predicting the disease at all — and still be clinically useless. This simulator renders a 2D clinical-biomarker cohort exactly like that, then lets you apply the Synthetic Minority Oversampling Technique (SMOTE) to generate interpolated synthetic minority patients, retrain a real logistic-regression classifier via gradient descent, and watch recall, precision and F1 respond as the decision-boundary line sweeps through the point cloud — with a threshold slider that shows the always-present trade-off between catching more real cases and raising more false alarms. Pan and zoom the scatter plane freely; a second live panel tracks the three metrics as bars so you can see the recall/precision trade-off at a glance.