How vishwesh-01/unbswhybrid works
Main Components
• Data Source (UNSW-NB15 Excel file)
• Data Preprocessing
– Categorical encoding (LabelEncoders)
– Scaling (MinMaxScaler)
– SMOTE oversampling (optional)
• Split (Train/Test)
• Model Training Pipeline
– CNN layers → LSTM layers → Dense layers
– Early stopping, class weights
• Artifacts
– cnn_lstm_unsw_model.h5
– scaler.pkl, preprocess_info.pkl
• Inference & MAC Spoofing Detection
Relationships & Data Flow
• Dataset → Preprocessing → SMOTE? → Split → Reshape
• Reshaped data → Model build & training → Evaluation → Save artifacts
• Artifacts + Incoming sample → Inference → MAC spoofing check → Output
Architectural Patterns
• Linear, pipelined ML workflow
• Modular separation: data-prep, training, inference
• Configurable parameters loaded at runtime (batch size, LR, etc.)
Key Technologies & Frameworks
• Python 3.11
• TensorFlow 2.18 (Keras API)
• scikit-learn, imbalanced-learn, pandas
• Jupyter Notebook / Python script entry points