Module 6: classification and decision trees

{
  "module": 6,
  "dataset": "UCI red wine quality",
  "rows": 1599,
  "chosen_model": "logistic",
  "threshold": 0.3,
  "selection_metric": "validation balanced accuracy",
  "test": {
    "model": "logistic",
    "n": 243,
    "tp": 19,
    "tn": 184,
    "fp": 30,
    "fn": 10,
    "precision": 0.3877551020408163,
    "recall": 0.6551724137931034,
    "specificity": 0.8598130841121495,
    "selection_rate": 0.20164609053497942,
    "fpr": 0.14018691588785046,
    "balanced_accuracy": 0.7574927489526264,
    "f1": 0.48717948717948717
  },
  "split_group_overlap": false,
  "label": "human score >=7"
}

Chemistry-identical rows stay in one split. No physical sample IDs exist; independence cannot be guaranteed.

Quality >=7 is a declared classroom label, not a safety/quality certification. Thresholds are selected on validation only.

Confusion counts, denominators, and rates are explicit. Feature importance and tree rules describe this fit, not causal effects.