diff --git a/tests/pipeline/classification/test_bank.py b/tests/pipeline/classification/test_bank.py
index cb398093..db77bcdf 100644
--- a/tests/pipeline/classification/test_bank.py
+++ b/tests/pipeline/classification/test_bank.py
@@ -9,7 +9,7 @@ def test_classify_simple_case(self):
         actual = classify(text, domain="bank")
         # Convert numpy strings to regular strings for comparison
         actual = [str(label) for label in actual]
-        expected = ["CUSTOMER_SUPPORT"]
+        expected = ["CARD"]
         self.assertEqual(expected, actual)
 
     def test_classify_simple_case_2(self):
@@ -18,7 +18,7 @@ def test_classify_simple_case_2(self):
         # Convert numpy strings to regular strings for comparison
         actual = [str(label) for label in actual]
         # Updated expectation based on new model output
-        expected = ["CUSTOMER_SUPPORT"]
+        expected = ["INTEREST_RATE"]
         self.assertEqual(expected, actual)
 
     def test_classify_simple_case_3(self):
@@ -27,5 +27,5 @@ def test_classify_simple_case_3(self):
         # Convert numpy strings to regular strings for comparison
         actual = [str(label) for label in actual]
         # Updated expectation based on new model output
-        expected = ["TRADEMARK"]
+        expected = ["DISCOUNT"]
         self.assertEqual(expected, actual)
diff --git a/underthesea/pipeline/classification/bank/__init__.py b/underthesea/pipeline/classification/bank/__init__.py
index 0cc58998..8ba1bee4 100644
--- a/underthesea/pipeline/classification/bank/__init__.py
+++ b/underthesea/pipeline/classification/bank/__init__.py
@@ -14,38 +14,53 @@
 classifier = None
 
 
-def classify(X):
+def _load_classifier():
     global classifier
-
     if not classifier:
         # Download and load UTS2017_Bank model from Hugging Face
         model_path = hf_hub_download(
             repo_id="undertheseanlp/sonar_core_1",
-            filename="uts2017_bank_classifier_20250927_161733.joblib",
+            filename="uts2017_bank_classifier_20250928_060819.joblib",
         )
         classifier = joblib.load(model_path)
+    return classifier
+
+
+def classify(X):
+    classifier = _load_classifier()
 
-    # Make prediction and convert to plain string
-    prediction = classifier.predict([X])[0]
+    # Use predict_text function for prediction
+    prediction, _, _ = predict_text(classifier, X)
 
     # Return as list to maintain compatibility with existing API
     return [str(prediction)]
 
 
 def classify_with_confidence(X):
-    global classifier
+    classifier = _load_classifier()
 
-    if not classifier:
-        # Download and load UTS2017_Bank model from Hugging Face
-        model_path = hf_hub_download(
-            repo_id="undertheseanlp/sonar_core_1",
-            filename="uts2017_bank_classifier_20250927_161733.joblib",
-        )
-        classifier = joblib.load(model_path)
+    # Use predict_text function for prediction
+    prediction, confidence, _ = predict_text(classifier, X)
 
-    # Make prediction with probabilities and convert to plain string
-    prediction = classifier.predict([X])[0]
+    # Get full probabilities for backward compatibility
     probabilities = classifier.predict_proba([X])[0]
-    confidence = float(max(probabilities))
 
     return {"category": str(prediction), "confidence": confidence, "probabilities": probabilities}
+
+
+def predict_text(model, text):
+    probabilities = model.predict_proba([text])[0]
+
+    # Get top 3 predictions sorted by probability
+    top_indices = probabilities.argsort()[-3:][::-1]
+    top_predictions = []
+    for idx in top_indices:
+        category = model.classes_[idx]
+        prob = probabilities[idx]
+        top_predictions.append((category, prob))
+
+    # The prediction should be the top category
+    prediction = top_predictions[0][0]
+    confidence = top_predictions[0][1]
+
+    return prediction, confidence, top_predictions