ML Model Performance
5-Model Ensemble Detection System • Real-time Metrics
88.3%
Detection Accuracy
Industry avg: 78%
90.7%
Precision
Low false positives
66.7%
Recall
Threat detection rate
76.8%
F1 Score
Best-in-class
Predictive Lead Time
Our ML models predict attacks before they happen. On average, PCDS alerts security teams 72 hours before an attack reaches execution phase.
72h
Average Lead Time
24h
Minimum Lead
168h
Maximum Lead
91%
Prediction Accuracy
5-Model Ensemble
LSTM Sequence Detector
Temporal Patterns
89.2%
accuracy
Random Forest Classifier
Feature-based
92.1%
accuracy
Isolation Forest
Anomaly Detection
87.5%
accuracy
Behavioral Analyzer
UEBA
91.8%
accuracy
DGA Detector (CNN)
Domain Analysis
94.7%
accuracy
Combined Ensemble Accuracy
88.3%
Active Threat Predictions
Credential theft in 48h
Ransomware deployment 24h
Data exfiltration attempt
1.9ms
Avg Latency
< 50ms target
598 eps
Throughput
events/second
5.3M+
Training Data
samples
2.8%
False Positive Rate
industry: 15%
0.927
AUC-ROC
excellent
vs. Industry Leaders
| Metric | PCDS | Darktrace | Vectra | Industry Avg |
|---|---|---|---|---|
| Detection Accuracy | 88.3% | ~85% | ~82% | 78% |
| Predictive Lead Time | 72 hours | N/A (reactive) | N/A (reactive) | N/A |
| False Positive Rate | 2.8% | ~12% | ~10% | 15% |
| MITRE ATT&CK Coverage | 26% | ~40% | ~35% | 20% |