diff --git a/archives/fnc2.log b/archives/fnc2.log
new file mode 100644
index 0000000..756a36e
--- /dev/null
+++ b/archives/fnc2.log
@@ -0,0 +1,27 @@
+📚 Loading data from Parquet file at '../data/sampled_fakenews_train.parquet'
+📚 Loading data from Parquet file at '../data/sampled_fakenews_valid.parquet'
+📚 Loading data from Parquet file at '../data/sampled_fakenews_test.parquet'
+🧮 Grouping into binary classes...
+🪙 Preprocessing text...
+🔍 Training models...
+
+📊 Logistic Regression Test Performance:
+              precision    recall  f1-score   support
+
+    Reliable       0.84      0.90      0.87     54706
+        Fake       0.80      0.70      0.75     30584
+
+    accuracy                           0.83     85290
+   macro avg       0.82      0.80      0.81     85290
+weighted avg       0.83      0.83      0.83     85290
+
+
+📊 Naïve Bayes Test Performance:
+              precision    recall  f1-score   support
+
+    Reliable       0.79      0.92      0.85     54706
+        Fake       0.79      0.57      0.67     30584
+
+    accuracy                           0.79     85290
+   macro avg       0.79      0.74      0.76     85290
+weighted avg       0.79      0.79      0.78     85290
diff --git a/requirements.txt b/requirements.txt
index 6e1abf8..0bfa32a 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -4,4 +4,8 @@ spacy
 nltk
 matplotlib
 pandarallel
-pyarrow
\ No newline at end of file
+pyarrow
+scikit-learn
+torch
+transformers
+accelerate
\ No newline at end of file
diff --git a/src/fnc1b.py b/src/fnc1b.py
index 796fecd..4e8bc4f 100644
--- a/src/fnc1b.py
+++ b/src/fnc1b.py
@@ -72,4 +72,24 @@ print(f"✅ Sample contains {len(df_sample):,} rows (expected {get_sample_size(t
 df_sample.to_csv(f"{sample_path}.csv", index=False)
 df_sample.to_parquet(f"{sample_path}.parquet", index=False)
 
-print(f"💾 Sample saved to '{sample_path}.csv' and '{sample_path}.parquet'.")
\ No newline at end of file
+print(f"💾 Sample saved to '{sample_path}.csv' and '{sample_path}.parquet'.")
+
+# Split to 80/10/10 and save as both CSV and Parquet
+train_size = int(len(df_sample) * 0.8)
+valid_size = int(len(df_sample) * 0.1)
+test_size = len(df_sample) - (train_size + valid_size)  # Ensure the sum is correct
+
+df_train = df_sample.iloc[:train_size]
+df_valid = df_sample.iloc[train_size:train_size + valid_size]
+df_test = df_sample.iloc[train_size + valid_size:]
+
+df_train.to_csv(f"{sample_path}_train.csv", index=False)
+df_valid.to_csv(f"{sample_path}_valid.csv", index=False)
+df_test.to_csv(f"{sample_path}_test.csv", index=False)
+
+df_train.to_parquet(f"{sample_path}_train.parquet", index=False)
+df_valid.to_parquet(f"{sample_path}_valid.parquet", index=False)
+df_test.to_parquet(f"{sample_path}_test.parquet", index=False)
+
+print(f"💾 Train/Valid/Test splits saved to '{sample_path}_train.csv', '{sample_path}_valid.csv', '{sample_path}_test.csv'.")
+
diff --git a/src/fnc2.py b/src/fnc2.py
new file mode 100644
index 0000000..b6bc04b
--- /dev/null
+++ b/src/fnc2.py
@@ -0,0 +1,74 @@
+import pandas as pd
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.linear_model import LogisticRegression
+from sklearn.model_selection import GridSearchCV
+from sklearn.naive_bayes import MultinomialNB
+from sklearn.metrics import classification_report
+
+# Load parquet first, fall back to CSV if not available
+def load_split(file_prefix, split_name):
+    try:
+        print(f"📚 Loading data from Parquet file at '{file_prefix}_{split_name}.parquet'")
+        return pd.read_parquet(f"{file_prefix}_{split_name}.parquet")
+    except FileNotFoundError:
+        print(f"❌ Error: Neither Parquet nor CSV file found at {file_prefix}_{split_name}.parquet or {file_prefix}_{split_name}.csv")
+        exit()
+    except:
+        print(f"🔄 Parquet file not found, loading from CSV at '{file_prefix}_{split_name}.csv'")
+        return pd.read_csv(f"{file_prefix}_{split_name}.csv")
+
+train = load_split("../data/sampled_fakenews", "train")
+val = load_split("../data/sampled_fakenews", "valid") 
+test = load_split("../data/sampled_fakenews", "test")
+
+# "Political" and "bias" may not be inherently fake, and "unknown" is neutral
+print("🧮 Grouping into binary classes...")
+fake_labels = {'fake', 'conspiracy', 'rumor', 'unreliable', 'junksci', 'hate', 'satire', 'clickbait'}
+for df in [train, val, test]:
+    df['label'] = df['type'].apply(lambda x: 1 if x in fake_labels else 0)
+
+print("🪙 Preprocessing text...")
+tfidf = TfidfVectorizer(max_features=5000)
+
+X_train = tfidf.fit_transform(train['processed_text'])
+X_val = tfidf.transform(val['processed_text'])
+X_test = tfidf.transform(test['processed_text'])
+
+y_train = train['label']
+y_val = val['label']
+y_test = test['label']
+
+print("🔍 Training models...")
+lr = LogisticRegression(C=10, max_iter=100, class_weight=None, random_state=42)
+lr.fit(X_train, y_train)
+
+nb = MultinomialNB(alpha=0.1, fit_prior=True)
+nb.fit(X_train, y_train)
+
+y_test_pred_lr = lr.predict(X_test)
+print("\n📊 Logistic Regression Test Performance:")
+print(classification_report(y_test, y_test_pred_lr, target_names=['Reliable', 'Fake']))
+
+y_test_pred_nb = nb.predict(X_test)
+print("\n📊 Naïve Bayes Test Performance:")
+print(classification_report(y_test, y_test_pred_nb, target_names=['Reliable', 'Fake']))
+
+# Function to perform hyperparameter tuning, not used in the final script
+def hyperparameter_tuning():
+    print("🔍 Hyperparameter tuning...")
+    param_grid_lr = {
+        'C': [0.1, 1, 10],
+        'max_iter': [100, 500, 1000],
+        'class_weight': ['balanced', None]
+    }
+    grid = GridSearchCV(LogisticRegression(), param_grid_lr, cv=3)
+    grid.fit(X_train, y_train)
+    print("✅ Best Logistic Regression Parameters:", grid.best_params_)
+
+    param_grid_nb = {
+        'alpha': [0.1, 0.5, 1.0, 2.0],
+        'fit_prior': [True, False]
+    }
+    grid_nb = GridSearchCV(MultinomialNB(), param_grid_nb, cv=3) 
+    grid_nb.fit(X_val, y_val)
+    print("✅ Best Naïve Bayes Parameters:", grid_nb.best_params_)