Sentinel XAI

Explainable Threat Detection · v2.4.1

Live

Security operations overview

Real-time detection with transparent model reasoning. Every alert comes with a human-readable explanation.

Active threats in feed

09

streaming live

Critical alerts open

03

3 need triage

Model accuracy

98.0%

+0.17pp · F1 0.968

Events / tick

520

-43 · Σ 16.0K

Traffic vs. detections
Rolling live window · updates as new telemetry arrives.
Events · 520Threats · 1
Attack type distribution
Share of the 9 most recent detections. Updates live.
  • Brute Force
    2 · 22.2%
  • SQL Injection
    1 · 11.1%
  • DDoS
    1 · 11.1%
  • Data Exfiltration
    1 · 11.1%
  • Port Scan
    1 · 11.1%
  • Malware C2
    1 · 11.1%
  • Phishing
    1 · 11.1%
  • Privilege Escalation
    1 · 11.1%
Live threat feed
9 recent events · tap any row to see the model explanation.
streaming
  • Brute Force185.220.101.42 → :22
    critical
    15:52:31
    97.0%active
  • SQL Injection92.118.161.18 → :443
    critical
    15:51:01
    94.0%investigating
  • DDoS203.0.113.88 → :80
    high
    15:49:31
    88.0%contained
  • Data Exfiltration10.0.3.41 → :8443
    critical
    15:48:01
    91.0%active
  • Port Scan41.220.77.9 → :0
    medium
    15:46:31
    76.0%contained
  • Malware C210.0.2.88 → :6667
    high
    15:45:01
    89.0%investigating
  • Phishing198.51.100.14 → :25
    medium
    15:43:31
    72.0%contained
  • Privilege Escalation10.0.4.44 → :389
    high
    15:42:01
    84.0%active
  • Brute Force172.58.12.9 → :443
    low
    15:40:31
    58.0%contained
Why was this flagged?· EVT-48291
Per-prediction explanation using SHAP and LIME
Brute Force

Prediction

THREAT

threshold 0.50 · TCP

Confidence

97.0%

Origin

185.220.101.42

RU · dest :22

Base value 0.12f(x) = 0.97
pushes toward threatpushes toward benign

Each row shows how much a feature shifted the prediction from the base value. Positive (red) contributions raise the threat score; negative (cyan) contributions lower it.

  • failed_login_attempts

    = 247 in 60s

    +0.38
  • source_ip_reputation

    = malicious (Tor exit)

    +0.24
  • auth_protocol

    = SSH

    +0.11
  • geo_risk_score

    = 0.89

    +0.09
  • user_history

    = first contact

    −0.06
  • time_of_day

    = 03:12 local

    +0.05
  • tls_fingerprint

    = known botnet

    +0.04
Global feature importance
Average absolute SHAP value across the model's training distribution.