How mazen14125/neural-network-for-phishing-email-detection works
Project type and purpose
• This is a machine-learning classification pipeline packaged in a Jupyter notebook.
• Purpose: ingest the Enron Email Dataset, preprocess text, train a Keras/TensorFlow neural network, and evaluate performance on phishing email detection.
File structure
• Top-level files:
• Neural Network for Phishing Email Detection.ipynb – contains all data-prep, modeling, and evaluation code.
• README.md – outlines dataset, preprocessing steps, model architecture, and results.
• No separate source folders, services, or configs—everything lives in the notebook.
Extract insights from the README
• Data preparation: tokenization, stemming, TF-IDF vectorization (scikit-learn)
• Model: multi-layer neural network (Keras/TensorFlow)
• Evaluation: accuracy, precision, recall (scikit-learn metrics)
• Visualization: Matplotlib & Seaborn
Main components
a. Data Source
• Enron Email Dataset (CSV files or database dump)
• External link (Kaggle)
b. Preprocessing Pipeline
• Tokenizer
• Stemmer
• TF-IDF Vectorizer
c. Neural Network Training
• Keras/TensorFlow model
• Layers (input, hidden, output)
• Training loop with loss and optimizer
d. Model Storage
• Saved model artifacts (HDF5 or SavedModel format)
e. Evaluation Module
• Metrics calculation (accuracy, precision, recall)
• Visualization (charts, confusion matrix)
f. Inference (optional)
• Load saved model to classify new emails
Architectural patterns and design principles
• Pipeline pattern: sequential data transformations
• Separation of concerns: data ingestion, preprocessing, training, evaluation
• Reproducibility: single notebook orchestrates all steps, but can be modularized into scripts
Technologies and libraries
• Python runtime
• Pandas & NumPy for data handling
• scikit-learn for TF-IDF and metrics
• Keras/TensorFlow for model definition and training
• Matplotlib & Seaborn for charts
Detail
• Exact sequence: raw emails → tokenization → stemming → vectorization → model training → metrics.
• Library dependencies on each stage.