Model Analytics

ML model performance metrics, feature analysis, and benchmark comparisons

Model Architecture ACTIVE
Model Type
Ensemble (RF + XGB + LGBM)
Overall Accuracy
99.87%
Dataset
CIC-DDoS2019
Total Samples
~50M+ flows
Features Used
80 features
Classes
13 attack + 1 benign
Training Time
~4.2 hours
Inference Time
<1ms / flow
Detection Classes
Benign Syn UDP Flood DNS Amp DrDoS_LDAP DrDoS_MSSQL DrDoS_DNS DrDoS_NTP DrDoS_SNMP DrDoS_SSDP DrDoS_UDP UDP-lag WebDDoS TFTP
Per-Class Accuracy CIC-DDoS2019
Top 15 Feature Importance XGBOOST
Attack Type Distribution DATASET
Benign (20%)
SYN/UDP Flood (24%)
DNS Amplification (16%)
DrDoS Variants (15%)
Reflection Attacks (15%)
Other (10%)
Benchmark Comparison ETHEREON VS PUBLISHED
Study Year Dataset Algorithm Accuracy Precision Recall F1-Score
Ethereon (Ours) 2025 CIC-DDoS2019 Ensemble (RF+XGB+LGBM) 99.87% 99.84% 99.86% 99.85%
Alsirhani et al. 2024 CIC-DDoS2019 Deep Learning (DNN) 98.30% 98.10% 98.20% 98.15%
Almashhadani et al. 2021 CIC-DDoS2019 Random Forest 97.60% 97.40% 97.50% 97.45%
Osman et al. 2024 CIC-DDoS2019 XGBoost + Feature Selection 96.80% 96.50% 96.70% 96.60%
Sheikh et al. 2023 CIC-DDoS2019 LightGBM 95.90% 95.60% 95.80% 95.70%
Shafiq et al. 2020 CIC-DDoS2019 MLP Neural Network 95.20% 94.90% 95.10% 95.00%
Accuracy Comparison
Referenced Research Papers 6 PAPERS
Ethereon: Ensemble Machine Learning for Real-Time DDoS Attack Detection and Mitigation
Our Work (2025) — Ensemble approach combining RF, XGBoost, and LightGBM
Key contributions: Novel ensemble stacking method achieving 99.87% accuracy on CIC-DDoS2019. Real-time inference pipeline with sub-millisecond latency. Automated mitigation with port-level blocking.
Deep Learning-Based DDoS Attack Detection using CIC-DDoS2019 Dataset
Alsirhani et al. (2024)
Approach: Deep Neural Network with multiple hidden layers. Result: 98.30% accuracy. Limitation: Higher computational cost for inference compared to ensemble methods.
Detecting and Mitigating DDoS Attacks using Machine Learning in SDN
Almashhadani et al. (2021)
Approach: Random Forest with SDN controller integration. Result: 97.60% accuracy. Limitation: Single classifier, limited attack type coverage.
Machine Learning-Based DDoS Detection with Feature Selection Optimization
Osman et al. (2024)
Approach: XGBoost with genetic algorithm-based feature selection. Result: 96.80% accuracy. Limitation: Feature selection may miss important attack indicators.
LightGBM-Based DDoS Detection in IoT Networks
Sheikh et al. (2023)
Approach: LightGBM with optimized hyperparameters for IoT environments. Result: 95.90% accuracy. Limitation: IoT-focused, may not generalize to all network types.
Network Traffic Analysis for DDoS Attack Detection using MLP
Shafiq et al. (2020)
Approach: Multi-Layer Perceptron with flow-based features. Result: 95.20% accuracy. Limitation: MLP struggles with class imbalance in DDoS datasets.