feat: exclude additional metadata columns from feature loading to prevent data leakage

app/training/evaluate.py

@@ -45,9 +45,10 @@ def load_features_csv(path: str | Path) -> tuple[np.ndarray, np.ndarray]:
 
     with open(path, "r", encoding="utf-8") as f:
         reader = csv.DictReader(f)
+        _EXCLUDE = {"file_path", "label_int", "duration_sec", "sample_rate"}
         feature_cols = [
             c for c in reader.fieldnames
-            if c not in ("file_path", "label_int")
+            if c not in _EXCLUDE
         ]
 
         for row in reader:

app/training/train_classifier.py

@@ -95,9 +95,12 @@ def train(
 
     with open(features_csv, "r", encoding="utf-8") as f:
         reader = csv.DictReader(f)
+        # duration_sec and sample_rate are metadata, not audio features —
+        # including them causes data leakage (duration correlates with source, not content)
+        _EXCLUDE = {"file_path", "label_int", "duration_sec", "sample_rate"}
         feature_cols = [
             c for c in reader.fieldnames
-            if c not in ("file_path", "label_int")
+            if c not in _EXCLUDE
         ]
 
     # ── Handle NaN/Inf ─────────────────────────────