Web_page_phishing_detection

This project focuses on detecting phishing websites using machine learning models. The dataset is sourced from Kaggle's Web Page Phishing Detection Dataset. The analysis, model building, and evaluation were conducted in a Jupyter Notebook.

🪼 Dataset

• Source: Web Page Phishing Detection Dataset on Kaggle
• Rows: 11,430
• Columns: 89
• Target Variable: status (Labels: phishing, legitimate)
• Balanced Dataset: 5,715 phishing and 5,715 legitimate samples

🪼 Exploratory Data Analysis (EDA)

• Analyzed categorical and numerical features
• Plotted histograms and correlation matrices
• Found no missing values

🪼 Feature Selection (PCA)

• Dropped non-relevant columns like url
• Standardized data with StandardScaler
• Applied PCA to select top 10 principal components
• Final feature set reduced to 35 columns

🪼 Models & Evaluation

🐳 Models Trained

Model	Accuracy
Random Forest	95.5%
Logistic Regression	91.4%
SVM (RBF Kernel)	93.6%
KNN (k=5)	92.9%
Neural Network	94.3%

🐳 Metrics Used

• Accuracy
• Precision
• Recall
• F1-Score
• Confusion Matrix

🐳 K-Fold Cross-Validation (5-Fold)

Random Forest consistently performed the best with ~95% accuracy.

🐳 Neural Network Architecture (Keras)

• Dense(64, ReLU) → Dropout(0.3)
• Dense(32, ReLU) → Dropout(0.2)
• Output: Dense(1, Sigmoid)
• Optimizer: Adam, Loss: Binary Crossentropy
• Best Accuracy: ~94.6% (Validation), 94.3% (Test)

🪼 Outlier Handling & Retraining

Method	Records After	Best Accuracy
IQR	2,970	Slight drop
Z-Score	9,195	~94.4% (RF)

🪼 Interactive Dashboard (Dash + Plotly)

Features:

• Histograms, Box Plots, Heatmaps
• Class-based Feature Comparison
• Model Performance Visualizations
• Confusion Matrices
• Neural Network Training Curves

🪼 Key Takeaways

• PCA effectively reduced dimensionality.
• Random Forest and Neural Network models performed best.
• Dash-based dashboard enhances interpretability and presentation.
• Outlier removal confirmed model robustness.

🪼 How to Run

1. Install libraries:

   pip install pandas numpy matplotlib seaborn scikit-learn keras xgboost dash
   

2. Run the notebook PhishingDetection.ipynb.

🪼 Author

Ananya P S