How marwenkouidhi/automatic-detection-of-email-phishing-attacks works
Main Components
• Data Sources / Datasets: Represent the two primary external datasets (Nazario and TREC phishing corpora) mentioned in the README.
• Preprocessing Module: Represented by the preprocess directory, which contains notebooks that load data, merge datasets, extract features, and prepare DataFrames for use by the models.
• Modeling Module: Represented by the modeling directory (noting the README refers to a “models” directory, but the file tree shows “modeling”). This module contains various notebooks, each corresponding to a different phishing detection approach:
• NLP-based email content classification
• Header-based abnormality detection
• Unwanted content (e.g., suspicious links) detection
• Environment / Configuration: The environment.yml file which defines the Conda environment and installed dependencies.
• (Optionally) Data Storage or Visualization: Even if implicit, you may wish to denote intermediate data storage (DataFrames), especially since the notebooks indicate data transformations.
Relationships and Interactions
• Optionally, indicate overall decision logic or integration where the output of each model might be aggregated to produce a final phishing/non-phishing decision.
Architectural Patterns and Design Principles
• Modular Design: The separation between the preprocessing stage and each individual model ensures modularity and easier testing or improvements.
• Pipeline Architecture: The system follows a clear pipeline structure from data ingestion, preprocessing, to model inference.
• Reproducibility: Using a Conda environment (specified in environment.yml) ensures the computational environment is reproducible.
Specific Technologies, Frameworks, or Libraries
• Although not detailed in every notebook, highlight that TensorFlow and BERT are used in at least one model for fake news detection as per one of the notebook names.
• NLP libraries for processing email content.
• Data manipulation tools (Pandas for DataFrame handling) are implied in the preprocessing scripts.