This is the official GitHub repository for the paper "ZENA: Enhanced Anomaly-Based IDS for Smart Homes Automation Through Multilayer Artificial Neural Network" submitted to the Journal of Communications and Networks.

ZENA is an enhanced anomaly-based Intrusion Detection System (IDS) designed specifically for smart home automation environments. The system employs a multilayer artificial neural network to detect anomalous behaviors and potential security threats in smart home networks.

ZENA implements a three-layer neural network architecture:

• Input Layer: 30 features
• First Hidden Layer: 128 neurons with ReLU activation
• Second Hidden Layer: 32 neurons with ReLU activation
• Output Layer: Binary classification (sigmoid activation)

The model uses momentum-based gradient descent optimization and supports multiple weight initialization methods (Xavier, He, random, zero).

• Multi-layer artificial neural network for anomaly detection
• GPU acceleration support (optional, via CuPy)
• Early stopping mechanism to prevent overfitting
• Support for customizable hyperparameters
• Model parameter saving and loading functionality
• Comprehensive evaluation metrics (Accuracy, F1-Score)

• Python 3.x
• NumPy >= 1.19.0
• Pandas >= 1.0.0
• scikit-learn >= 0.23.0

• CuPy (for GPU acceleration)
  • For CUDA 11.x: pip install cupy-cuda11x
  • For CUDA 12.x: pip install cupy-cuda12x
  • If CuPy is not installed, the model will automatically use CPU (NumPy)

pip install -r requirements.txt

Train the ZENA model with default parameters:

python train.py

Train with custom parameters:

python train.py \
    --model_name ZENA \
    --input_size 30 \
    --hidden_size1 128 \
    --hidden_size2 32 \
    --learning_rate 0.03 \
    --iterations 100000 \
    --early_stopping_rounds 5000 \
    --tolerance 1e-4 \
    --init_method xavier \
    --momentum 0.9

Train with GPU acceleration (requires CuPy):

python train.py --use_gpu

from model import ZENA
import numpy as np

# Initialize the model (CPU)
model = ZENA(
    input_size=30,
    hidden_size1=128,
    hidden_size2=32,
    learning_rate=0.03,
    iterations=100000,
    early_stopping_rounds=5000,
    tolerance=1e-4,
    init_method='xavier',
    momentum=0.9
)

# Initialize the model with GPU acceleration
model_gpu = ZENA(
    input_size=30,
    hidden_size1=128,
    hidden_size2=32,
    learning_rate=0.03,
    iterations=100000,
    early_stopping_rounds=5000,
    tolerance=1e-4,
    init_method='xavier',
    momentum=0.9,
    use_gpu=True  # Enable GPU acceleration
)

# Train the model
train_loss_history, val_loss_history = model.fit(X_train, Y_train, X_val, Y_val)

# Make predictions
predictions = model.predict(X_test)

# Save model parameters
model.save_params('params/ZENA_model.pkl')

# Load model parameters
model.load_params('params/ZENA_model.pkl')

The model supports optional GPU acceleration through CuPy. If use_gpu=True is specified but CuPy is not installed, the model will automatically fall back to CPU (NumPy) with a warning message.

Installing CuPy:

• For NVIDIA GPUs with CUDA 11.x: pip install cupy-cuda11x
• For NVIDIA GPUs with CUDA 12.x: pip install cupy-cuda12x
• For ROCm (AMD GPUs): pip install cupy-rocm-5-0 (or appropriate version)

GPU acceleration can significantly speed up training, especially for large datasets or many iterations.

The model expects CSV files with the following format:

• Last column: Labels (0 for normal, 1 for anomaly)
• Other columns: Feature values (30 features by default)

The data will be automatically split into training (80%), validation (10%), and test (10%) sets.

ZENA/
├── model.py              # ZENA model implementation
├── train.py              # Training script
├── requirements.txt      # Python dependencies
├── README.md            # This file
├── ZENA_data.csv        # Training dataset
├── params/              # Saved model parameters
└── images/              # Figures and visualizations