Hybrid Deep Learning Architecture
CNN Layer
Convolutional Neural Network for spatial feature extraction from network traffic patterns. Detects local correlations in input features.
Conv1D(64, 3) → Conv1D(128, 3) → MaxPool → Flatten
MLP Layer
Multi-Layer Perceptron for deep feature transformation. Learns complex non-linear relationships between extracted features.
Dense(256) → Dropout(0.3) → Dense(128) → Dense(64)
LSTM Layer
Long Short-Term Memory for temporal sequence analysis. Captures sequential dependencies in network connections.
LSTM(128) → Dropout(0.2) → LSTM(64) → Dense(32)
CONCATENATION
Features from all three branches are concatenated into a unified representation → Dense(128) → Dropout(0.3) → Dense(5) → Softmax
Network Traffic Features Input
Enter network traffic features to classify the connection using the hybrid CNN+MLP+LSTM model.
Prediction Result
Enter features and run prediction to see results
Batch Traffic Analysis
Simulate batch prediction on multiple network connections sourced from KDD-Cup'99 dataset patterns.
Model Performance Metrics
Confusion Matrix (KDD-Cup'99)
Accuracy Comparison
Dataset Reference
KDD-Cup'99
4 Attack Classes • 41 Features
Normal, DOS, Probe, U2R, R2L
NSL-KDD
24 Attack Types • 4 Classes
Improved version without redundant records
UNSW-NB15
9 Attack Classes • 49 Features
Modern dataset with contemporary attack types
Hybrid Model Info
CNN + MLP + LSTM Concatenated
Combines spatial, deep non-linear, and temporal feature extraction for superior accuracy over individual models
Research: Novel Cyber Attack Detection Using a Hybrid Deep Learning Model
Credits: Preethika N C • Swetha Nachimuthu • Subiksha T
|
Dataset: KDD-Cup'99 / NSL-KDD / UNSW-NB15