models/metadata.json

{
  "model": "DecisionTree(max_depth=2)",
  "sklearn_version": "1.9.0",
  "created": "2026-07-01",
  "random_state": 42,
  "feature_order": [
    "Time_spent_Alone",
    "Social_event_attendance",
    "Going_outside",
    "Friends_circle_size",
    "Post_frequency",
    "Stage_fear",
    "Drained_after_socializing"
  ],
  "numeric_features": [
    "Time_spent_Alone",
    "Social_event_attendance",
    "Going_outside",
    "Friends_circle_size",
    "Post_frequency"
  ],
  "binary_features": [
    "Stage_fear",
    "Drained_after_socializing"
  ],
  "class_labels": [
    "Extrovert",
    "Introvert"
  ],
  "positive_class": "Introvert",
  "decision_threshold": 0.5,
  "n_rows_total": 2512,
  "n_train": 2009,
  "n_test": 503,
  "reported_test_metrics": {
    "n_test": 503,
    "accuracy": 0.9145,
    "precision_introvert": 0.8793,
    "recall_introvert": 0.9315,
    "f1_introvert": 0.9047,
    "roc_auc": 0.9242,
    "confusion_matrix": {
      "labels": [
        "Extrovert",
        "Introvert"
      ],
      "matrix": [
        [
          256,
          28
        ],
        [
          15,
          204
        ]
      ]
    }
  },
  "cv_metrics_depth2": {
    "cv": "stratified 5-fold on training split",
    "accuracy_mean": 0.9298,
    "accuracy_std": 0.0075,
    "roc_auc_mean": 0.9339
  },
  "note": "The serialized artifact (pipeline.joblib) is REFIT on 100% of rows (train + test) for deployment. The reported_test_metrics come from the held-out 80/20 test split and are the honest performance estimate; cv_metrics_depth2 come from stratified 5-fold CV on the training portion only."
}