train_spam_model.py

# -----------------------------
# 1️⃣ Import des librairies
# -----------------------------
import pandas as pd
import re
import joblib
import nltk
from nltk.corpus import stopwords
from nltk.stem import PorterStemmer
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score, roc_auc_score
from imblearn.over_sampling import SMOTE

# Télécharger stopwords si nécessaire
nltk.download('stopwords')

# -----------------------------
# 2️⃣ Prétraitement des messages
# -----------------------------
stop_words = set(stopwords.words('english'))
stemmer = PorterStemmer()

def preprocess_message(text):
    if pd.isna(text):
        return ""
    text = text.lower()
    text = re.sub(r'http\S+|www\S+', '', text)  # supprimer URLs
    text = re.sub(r'\S+@\S+', '', text)         # supprimer emails
    text = re.sub(r'\+?\d[\d -]{8,}\d', '', text)  # supprimer numéros
    text = re.sub(r'\d+', '', text)                # supprimer chiffres
    text = re.sub(r'[^a-z\s!/+>]', '', text)      # garder ponctuation utile spam
    words = [stemmer.stem(word) for word in text.split() if word not in stop_words]
    return " ".join(words)

# -----------------------------
# 3️⃣ Charger les données
# -----------------------------
# data doit avoir les colonnes "Message" et "Category" ('spam'/'ham')
data = pd.read_csv("data.csv")
data['cleaned'] = data['Message'].apply(preprocess_message)

X = data['cleaned']
y = data['Category']

# Split train/test stratifié
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42, stratify=y
)

# -----------------------------
# 4️⃣ Vectorisation TF-IDF
# -----------------------------
tfidf = TfidfVectorizer(
    max_features=5000,
    min_df=2,
    max_df=0.95,
    ngram_range=(1,2),
    token_pattern=r'(?u)\b\w+\b|[!/+>]'  # capture mots et ponctuations importantes
)
X_train_tfidf = tfidf.fit_transform(X_train)
X_test_tfidf = tfidf.transform(X_test)

# -----------------------------
# 5️⃣ Équilibrage des classes avec SMOTE
# -----------------------------
smote = SMOTE(random_state=42)
X_train_balanced, y_train_balanced = smote.fit_resample(X_train_tfidf, y_train)

# -----------------------------
# 6️⃣ Entraînement du modèle Logistic Regression
# -----------------------------
model = LogisticRegression(random_state=42, max_iter=1000)
model.fit(X_train_balanced, y_train_balanced)

# -----------------------------
# 7️⃣ Évaluation rapide
# -----------------------------
y_pred = model.predict(X_test_tfidf)
print("Classification Report:\n", classification_report(y_test, y_pred))
print("Matrice de confusion:\n", confusion_matrix(y_test, y_pred))
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy:.4f}")

if hasattr(model, 'predict_proba'):
    y_test_binary = (y_test == 'spam').astype(int)
    auc = roc_auc_score(y_test_binary, model.predict_proba(X_test_tfidf)[:,1])
    print(f"AUC-ROC: {auc:.4f}")

# -----------------------------
# 8️⃣ Sauvegarder modèle et TF-IDF
# -----------------------------
joblib.dump(model, "spam_model.pkl")
joblib.dump(tfidf, "tfidf_vectorizer.pkl")
print("✅ Modèle Logistic Regression et TF-IDF vectorizer sauvegardés avec succès !")