Scribd
Upload
28 views8 pages

Gmail Model - Ipynb - Colab

The document outlines the installation and implementation of a spam email classifier using Python libraries such as TensorFlow, scikit-learn, and imbalanced-learn. It includes data preprocessing, feature extraction, model training, and evaluation, culminating in a prediction function for classifying emails as spam or ham. The model is saved for future use, and example predictions are provided for demonstration.

Uploaded by

yaskalai1602
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
28 views8 pages

Gmail Model - Ipynb - Colab

The document outlines the installation and implementation of a spam email classifier using Python libraries such as TensorFlow, scikit-learn, and imbalanced-learn. It includes data preprocessing, feature extraction, model training, and evaluation, culminating in a prediction function for classifying emails as spam or ham. The model is saved for future use, and example predictions are provided for demonstration.

Uploaded by

yaskalai1602
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

!

pip install tldextract

Collecting tldextract
Downloading [Link] (11 kB)
Requirement already satisfied: idna in /usr/local/lib/python3.11/dist-packages (
Requirement already satisfied: requests>=2.1.0 in /usr/local/lib/python3.11/dist
Collecting requests-file>=1.4 (from tldextract)
Downloading requests_file-[Link] (1.7 kB)
Requirement already satisfied: filelock>=3.0.8 in /usr/local/lib/python3.11/dist
Requirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python
Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.11/d
Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.11/d
Downloading [Link] (107 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 107.4/107.4 kB 4.1 MB/s eta 0
Downloading requests_file-[Link] (4.2 kB)
Installing collected packages: requests-file, tldextract
Successfully installed requests-file-2.1.0 tldextract-5.3.0

!pip install scikit-learn imbalanced-learn matplotlib seaborn

import numpy as np
import pandas as pd
import re
import [Link] as plt
import seaborn as sns
import pickle
from sklearn.model_selection import train_test_split
from [Link] import StandardScaler
from [Link] import classification_report, confusion_matrix, roc_auc_score
from imblearn.over_sampling import RandomOverSampler
import tensorflow as tf
from [Link] import Model
from [Link] import (Input, Embedding, Conv1D, MaxPooling1D,
Bidirectional, LSTM, Dense, Dropout,
BatchNormalization, GlobalMaxPooling1D,
Concatenate)
from [Link] import Adam
from [Link] import EarlyStopping, ReduceLROnPlateau, ModelCheckpoint
from [Link] import Tokenizer
from [Link] import pad_sequences
from [Link] import l2

# Load dataset
df = pd.read_csv('/content/drive/MyDrive/kaggle_datasets/email_dataset/spam_Emails_data.csv
df = df[['text', 'label']].dropna()

# Preprocessing pipeline
def preprocess_text(text):
text = str(text).lower()
# Remove HTML tags
text = [Link](r'<[^>]+>', '', text)
# Replace URLs with 'URL'
text = [Link](r'https?://\S+|www\.\S+', 'URL', text)
# Replace email addresses with 'EMAIL'
text = [Link](r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b', 'EMAIL', text)
# Replace currency symbols
text = [Link](r'[$€£¥]\d+\.?\d*', 'CURRENCY', text)
# Normalize common obfuscations
text = [Link](r'v[i1!][a@]gr[a@]', 'viagra', text)
text = [Link](r'fr[e3][e3]', 'free', text)
# Replace multiple spaces with single space
text = [Link](r'\s+', ' ', text).strip()
return text

df['text'] = df['text'].apply(preprocess_text)

# Encode labels
df['label'] = df['label'].map({'Ham': 0, 'Spam': 1})

# Expanded keyword lists


SPAM_KEYWORDS = ['free', 'win', 'prize', 'offer', 'lottery', 'claim', 'exclusive', 'discount
'deal', 'bonus', 'gift', 'reward', 'limited', 'special', 'cash', 'money',
'save', 'buy', 'shop']
URGENCY_KEYWORDS = ['urgent', 'now', 'immediately', 'act', 'last', 'expire', 'deadline',
'final', 'today', 'quick', 'hurry']
PHISHING_KEYWORDS = ['verify', 'login', 'account', 'password', 'secure', 'update', 'confirm
'alert', 'suspended']
SCAM_KEYWORDS = ['inheritance', 'bank', 'transfer', 'funds', 'payment', 'deposit', 'million
'billion']
CALL_TO_ACTION = ['click here', 'visit now', 'call now', 'apply now', 'get now']

# Feature extraction
def extract_features(text):
features = [Link](20) # Increased to accommodate new features
text = str(text)
# Basic features
features[0] = len(text)
features[1] = [Link]('!')
features[2] = [Link]('?')
features[3] = [Link]('$')
features[4] = [Link]('@')
# Keyword counts
features[5] = sum([Link]().count(kw) for kw in SPAM_KEYWORDS)
features[6] = sum([Link]().count(kw) for kw in URGENCY_KEYWORDS)
features[7] = sum([Link]() for c in text) / max(1, len(text))
features[8] = sum([Link]() for c in text) / max(1, len(text))
features[9] = len([Link](r'URL', text))
features[10] = len([Link](r'EMAIL', text))
features[11] = len([Link](r'\b\d{5,}\b', text)) # Long numbers
features[12] = len([Link]()) # Word count
features[13] = len(set([Link]())) / max(1, len([Link]())) # Unique word ratio
features[14] = 1 if 'attachment' in [Link]() else 0
# New features
features[15] = len([Link](r'[\U0001F600-\U0001F64F\U0001F300-\U0001F5FF]', text)) #
features[16] = 1 if any(kw in [Link]() for kw in ['noreply', 'admin', 'support']) e
features[17] = features[5] / max(1, len([Link]())) # Spam keyword density
features[18] = len([Link](r'[*#~\^]', text)) / max(1, len(text)) # Special characte
features[19] = sum([Link]().count(phrase) for phrase in CALL_TO_ACTION) # Call-to-a
return features

X_num = [Link]([extract_features(text) for text in df['text']])


y = df['label'].values
# Scale features
scaler = StandardScaler()
X_num = scaler.fit_transform(X_num)

# Tokenization
max_words = 20000
max_len = 256 # Optimized for email length

tokenizer = Tokenizer(num_words=max_words, oov_token='<OOV>')


tokenizer.fit_on_texts(df['text'])
sequences = tokenizer.texts_to_sequences(df['text'])
X_text = pad_sequences(sequences, maxlen=max_len, padding='post', truncating='post')

# Class balancing
sampler = RandomOverSampler(random_state=42)
X_num, y = sampler.fit_resample(X_num, y)
X_text = [Link]([X_text[i] for i in sampler.sample_indices_])

# Split data
X_text_train, X_text_test, X_num_train, X_num_test, y_train, y_test = train_test_split(
X_text, X_num, y, test_size=0.2, random_state=42
)

X_text_train, X_text_val, X_num_train, X_num_val, y_train, y_val = train_test_split(


X_text_train, X_num_train, y_train, test_size=0.2, random_state=42
)

# Input layers
text_input = Input(shape=(max_len,), name='text_input')
num_input = Input(shape=(X_num.shape[1],), name='num_input')

# Text processing branch


x = Embedding(max_words, 128)(text_input)
x = Conv1D(64, 5, activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = MaxPooling1D(2)(x)
x = Conv1D(128, 3, activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = MaxPooling1D(2)(x)
x = Bidirectional(LSTM(64, return_sequences=True))(x)
x = GlobalMaxPooling1D()(x)
x = Dropout(0.5)(x)

# Numerical features branch


y = Dense(64, activation='relu')(num_input)
y = BatchNormalization()(y)
y = Dropout(0.3)(y)

# Combined model
combined = Concatenate()([x, y])
z = Dense(64, activation='relu')(combined)
z = Dropout(0.3)(z)
output = Dense(1, activation='sigmoid')(z)

model = Model(inputs=[text_input, num_input], outputs=output)

[Link](
optimizer=Adam(learning_rate=0.001),
loss='binary_crossentropy',
metrics=['accuracy',
[Link](name='precision'),
[Link](name='recall'),
[Link](name='auc')]
)

callbacks = [
EarlyStopping(monitor='val_auc', patience=5, mode='max', restore_best_weights=True),
ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=2, min_lr=1e-6),
ModelCheckpoint(filepath='best_gmail_spam_model.h5', monitor='val_auc', save_best_only=T
]

history = [Link](
[X_text_train, X_num_train], y_train,
validation_data=([X_text_val, X_num_val], y_val),
epochs=10, # Increased for better convergence
batch_size=64, # Slightly smaller for better gradient updates
callbacks=callbacks,
verbose=1
)

# Predictions
y_pred = ([Link]([X_text_test, X_num_test]) > 0.5).astype(int)

# Metrics
print(classification_report(y_test, y_pred, target_names=['Ham', 'Spam']))
print(f"AUC-ROC: {roc_auc_score(y_test, y_pred):.4f}")

# Confusion matrix
cm = confusion_matrix(y_test, y_pred)
[Link](cm, annot=True, fmt='d', cmap='Blues',
xticklabels=['Ham', 'Spam'], yticklabels=['Ham', 'Spam'])
[Link]('Predicted')
[Link]('Actual')
[Link]()

[Link]('gmail_spam_model.h5')
with open('[Link]', 'wb') as f:
[Link](tokenizer, f)
with open('[Link]', 'wb') as f:
[Link](scaler, f)

# Prediction script
def predict_email(text):
"""Predict whether an email is spam or ham"""
PREDICTION_THRESHOLD = 0.5
# Preprocess text
cleaned_text = preprocess_text(text)

# Prepare text features


sequence = tokenizer.texts_to_sequences([cleaned_text])
padded_sequence = pad_sequences(sequence, maxlen=256, padding='post', truncating='post'

# Prepare numerical features


num_features = extract_features(text)
scaled_features = [Link]([num_features])

# Make prediction
prediction_prob = [Link]([padded_sequence, scaled_features])[0][0]
is_spam = prediction_prob > PREDICTION_THRESHOLD
# Generate human-readable result
result = {
'text_sample': text[:100] + "..." if len(text) > 100 else text,
'prediction': 'SPAM' if is_spam else 'HAM',
'probability': float(prediction_prob),
'confidence': 'HIGH' if (prediction_prob > 0.8 or prediction_prob < 0.2) else 'MEDIU
}

# Explain key factors


suspicious_factors = []
if num_features[1] > 3:
suspicious_factors.append(f"Multiple (!) marks ({int(num_features[1])})")
if num_features[3] > 0:
suspicious_factors.append("Contains dollar signs")
if num_features[5] > 0:
suspicious_factors.append(f"Spam keywords ({int(num_features[5])})")
if num_features[9] > 0:
suspicious_factors.append(f"Contains URLs ({int(num_features[9])})")
if num_features[7] > 0.3:
suspicious_factors.append(f"Excessive uppercase ({num_features[7]:.1%})")
if num_features[15] > 0:
suspicious_factors.append(f"Contains emojis ({int(num_features[15])})")
if num_features[16] > 0:
suspicious_factors.append("Suspicious sender pattern")
if num_features[19] > 0:
suspicious_factors.append(f"Call-to-action phrases ({int(num_features[19])})")

if suspicious_factors:
result['suspicious_factors'] = suspicious_factors

return result

# Example usage
test_emails = [
"WIN A FREE iPhone! Click now: [Link]
"Hi John, just following up about our meeting tomorrow",
"URGENT: Your account has been compromised. Verify now! $$$",
"Please find attached the quarterly report",
"Congratulations! You've won a $1000 gift card! Click: [Link]
"Dear Sir, you have inherited $5 million from a relative. Contact us immediately!"
]

print("="*60)
print("GMAIL SPAM CLASSIFIER PREDICTIONS")
print("="*60)

for email in test_emails:


prediction = predict_email(email)
print(f"\nEmail: {prediction['text_sample']}")
print(f"Prediction: {prediction['prediction']} (Probability: {prediction['probability']
print(f"Confidence: {prediction['confidence']}")
if 'suspicious_factors' in prediction:
print("Suspicious factors detected:")
for factor in prediction['suspicious_factors']:
print(f"- {factor}")
print("-"*60)
Requirement already satisfied: scikit-learn in /usr/local/lib/python3.11/dist-pa
Requirement already satisfied: imbalanced-learn in /usr/local/lib/python3.11/dis
Requirement already satisfied: matplotlib in /usr/local/lib/python3.11/dist-pack
Requirement already satisfied: seaborn in /usr/local/lib/python3.11/dist-package
Requirement already satisfied: numpy>=1.19.5 in /usr/local/lib/python3.11/dist-p
Requirement already satisfied: scipy>=1.6.0 in /usr/local/lib/python3.11/dist-pa
Requirement already satisfied: joblib>=1.2.0 in /usr/local/lib/python3.11/dist-p
Requirement already satisfied: threadpoolctl>=3.1.0 in /usr/local/lib/python3.11
Requirement already satisfied: sklearn-compat<1,>=0.1 in /usr/local/lib/python3.
Requirement already satisfied: contourpy>=1.0.1 in /usr/local/lib/python3.11/dis
Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.11/dist-pa
Requirement already satisfied: fonttools>=4.22.0 in /usr/local/lib/python3.11/di
Requirement already satisfied: kiwisolver>=1.3.1 in /usr/local/lib/python3.11/di
Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.11/dist
Requirement already satisfied: pillow>=8 in /usr/local/lib/python3.11/dist-packag
Requirement already satisfied: pyparsing>=2.3.1 in /usr/local/lib/python3.11/dis
Requirement already satisfied: python-dateutil>=2.7 in /usr/local/lib/python3.11
Requirement already satisfied: pandas>=1.2 in /usr/local/lib/python3.11/dist-pac
Requirement already satisfied: pytz>=2020.1 in /usr/local/lib/python3.11/dist-pa
Requirement already satisfied: tzdata>=2022.7 in /usr/local/lib/python3.11/dist-
Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.11/dist-packag
Epoch 1/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 188ms/step - accuracy: 0.9242 - auc: 0.9702
Epoch 1: val_auc improved from -inf to 0.99826, saving model to best_gmail_spam_m
WARNING:absl:You are saving your model as an HDF5 file via `[Link]()` or `ke
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 412s 199ms/step - accuracy: 0.9242 - auc: 0.97
Epoch 2/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 189ms/step - accuracy: 0.9895 - auc: 0.9987
Epoch 2: val_auc did not improve from 0.99826
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 463s 210ms/step - accuracy: 0.9895 - auc: 0.99
Epoch 3/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 189ms/step - accuracy: 0.9938 - auc: 0.9993
Epoch 3: val_auc did not improve from 0.99826
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 423s 200ms/step - accuracy: 0.9938 - auc: 0.99
Epoch 4/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 189ms/step - accuracy: 0.9966 - auc: 0.9996
Epoch 4: val_auc did not improve from 0.99826
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 441s 200ms/step - accuracy: 0.9966 - auc: 0.99
Epoch 5/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 189ms/step - accuracy: 0.9984 - auc: 0.9998
Epoch 5: val_auc did not improve from 0.99826
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 459s 209ms/step - accuracy: 0.9984 - auc: 0.99
Epoch 6/10
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 0s 189ms/step - accuracy: 0.9995 - auc: 0.9999
Epoch 6: val_auc did not improve from 0.99826
2044/2044 ━━━━━━━━━━━━━━━━━━━━ 442s 209ms/step - accuracy: 0.9995 - auc: 0.99
1277/1277 ━━━━━━━━━━━━━━━━━━━━ 32s 25ms/step
precision recall f1-score support

Ham 0.98 0.99 0.99 20581


Spam 0.99 0.98 0.99 20283

accuracy 0.99 40864


macro avg 0.99 0.99 0.99 40864
weighted avg 0.99 0.99 0.99 40864
AUC-ROC: 0.9862

WARNING:absl:You are saving your model as an HDF5 file via `[Link]()` or `ke
============================================================
GMAIL SPAM CLASSIFIER PREDICTIONS
============================================================
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 323ms/step

Email: WIN A FREE iPhone! Click now: [Link]


Prediction: SPAM (Probability: 81.72%)
Confidence: HIGH
Suspicious factors detected:
- Multiple (!) marks (4)
- Spam keywords (4)
------------------------------------------------------------
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 30ms/step

Email: Hi John, just following up about our meeting tomorrow


Prediction: HAM (Probability: 0.79%)
Confidence: HIGH
Suspicious factors detected:
- Spam keywords (1)
------------------------------------------------------------
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 29ms/step

Email: URGENT: Your account has been compromised. Verify now! $$$
Prediction: SPAM (Probability: 98.70%)
Confidence: HIGH
Suspicious factors detected:
- Contains dollar signs
------------------------------------------------------------
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 29ms/step
Email: Please find attached the quarterly report
Prediction: HAM (Probability: 0.54%)
Confidence: HIGH
------------------------------------------------------------
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 32ms/step

Email: Congratulations! You've won a $1000 gift card! Click: [Link]


Prediction: HAM (Probability: 43.60%)
Confidence: MEDIUM
Suspicious factors detected:
- Contains dollar signs
Spam keywords (2)

You might also like