Fake Website Detector Using Machine Learning – Real-Time Phishing Protection

1. Introduction

The Fake Website Detector Using Machine Learning is a Python-based cybersecurity project that detects fraudulent or phishing websites by analysing their structural, lexical, and content-based characteristics. The system uses machine learning algorithms to classify websites as Legitimate or Fake based on extracted URL features and HTML content attributes. By identifying malicious sites before users interact with them, the system helps prevent online scams, data theft, and phishing attacks. This solution can be integrated into browsers or used as a standalone application to enhance internet safety and user trust.

2. Existing System vs Proposed System

3. Working

1. Data Collection: Legitimate & phishing URLs (UCI, PhishTank).
2. Feature Extraction: URL length, “@”, subdomain count, HTTPS usage, domain age, etc.
3. Model Training: Train ML models to classify Legitimate vs Phishing.
4. Prediction Phase: User inputs URL → features analyzed → authenticity prediction.
5. Result Display: Safe/Suspicious/Fake with confidence score.

4. Technology Stack

• Language: Python
• Libraries: Scikit-learn, Pandas, NumPy, Flask, Regex, Matplotlib
• Algorithms: Random Forest, Logistic Regression, Decision Tree
• Dataset: PhishTank, UCI ML Repository (public datasets)
• Interface: Flask web UI for URL input & visualization
• Storage: CSV/SQLite for datasets & prediction logs

5. Modules

6. Advantages

• Automatic phishing detection in real time.
• ML outperforms static blacklists.
• Fast and lightweight on low-resource systems.
• Prevents credential theft & enhances browsing safety.
• Easy integration with browsers or security frameworks.

7. Applications

• Web browser security extensions for phishing protection.
• Banking & e-commerce fraud detection.
• Cybersecurity training in academic/corporate settings.
• Enterprise website monitoring & content filtering.

Python Integration Sketch (Flask + Scikit-learn)

# 1) Training
X, y = build_features(dataset_csv)     # lexical/domain features
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.2, stratify=y)
model = RandomForestClassifier(n_estimators=300, n_jobs=-1).fit(X_tr, y_tr)
print(classification_report(y_te, model.predict(X_te)))

# 2) Inference API
app = Flask(__name__)
@app.route('/predict', methods=['POST'])
def predict():
    url = request.json['url']
    feats = extract_features(url)      # length, @, subdomains, https, age, etc.
    proba = model.predict_proba([feats])[0]
    label_idx = int(proba.argmax())
    label = ['Safe','Phishing'][label_idx]
    score = float(proba[label_idx])
    log_to_sqlite(url, label, score)
    verdict = 'Suspicious' if 0.45 < score < 0.65 else label
    return jsonify({"verdict": verdict, "score": round(score,3)})

# 3) Flask UI posts to /predict and renders a badge + chart
              

What You Get