detecting-anomalous-authentication-patterns（检测异常身份验证模式）

Detecting Anomalous Authentication Patterns

When to Use

• Security operations needs to identify compromised accounts from authentication log analysis
• Implementing impossible travel detection to flag geographically inconsistent logins
• Detecting brute force, password spraying, and credential stuffing attacks in real time
• Building behavioral baselines for users to identify deviations indicating account compromise
• Correlating authentication anomalies with threat intelligence for lateral movement detection
• Investigating alerts from SIEM or IdP for suspicious sign-in activity

Do not use for static rule-based alerting on single failed logins; anomaly detection requires statistical baselines across time and entity dimensions to reduce false positives.

Prerequisites

• Authentication log sources (Azure AD/Entra ID sign-in logs, Okta system logs, Active Directory event logs 4624/4625/4648/4768/4771)
• SIEM platform (Splunk, Microsoft Sentinel, Elastic SIEM) with at least 90 days of baseline data
• GeoIP database for location-based anomaly detection (MaxMind GeoLite2 or IP2Location)
• Python 3.9+ with pandas, scikit-learn, and scipy for custom analytics
• User identity context (department, role, typical work hours, location)

Workflow

Step 1: Collect and Normalize Authentication Logs

Aggregate authentication events from all identity sources:

import pandas as pd
import json
from datetime import datetime, timedelta
from collections import defaultdict

# Parse authentication logs from multiple sources
def normalize_auth_logs(log_source, raw_logs):
    """Normalize authentication events to a common schema."""
    normalized = []

    for event in raw_logs:
        if log_source == "azure_ad":
            normalized.append({
                "timestamp": event["createdDateTime"],
                "user": event["userPrincipalName"],
                "source_ip": event["ipAddress"],
                "location": {
                    "city": event.get("location", {}).get("city"),
                    "state": event.get("location", {}).get("state"),
                    "country": event.get("location", {}).get("countryOrRegion"),
                    "lat": event.get("location", {}).get("geoCoordinates", {}).get("latitude"),
                    "lon": event.get("location", {}).get("geoCoordinates", {}).get("longitude")
                },
                "result": "success" if event["status"]["errorCode"] == 0 else "failure",
                "failure_reason": event["status"].get("failureReason", ""),
                "app": event.get("appDisplayName", "Unknown"),
                "device": event.get("deviceDetail", {}).get("operatingSystem", "Unknown"),
                "browser": event.get("deviceDetail", {}).get("browser", "Unknown"),
                "mfa_result": event.get("authenticationDetails", [{}])[0].get("succeeded", None),
                "risk_level": event.get("riskLevelDuringSignIn", "none"),
                "client_app": event.get("clientAppUsed", "Unknown"),
                "source": "azure_ad"
            })
        elif log_source == "okta":
            normalized.append({
                "timestamp": event["published"],
                "user": event["actor"]["alternateId"],
                "source_ip": event["client"]["ipAddress"],
                "location": {
                    "city": event["client"].get("geographicalContext", {}).get("city"),
                    "state": event["client"].get("geographicalContext", {}).get("state"),
                    "country": event["client"].get("geographicalContext", {}).get("country"),
                    "lat": event["client"].get("geographicalContext", {}).get("geolocation", {}).get("lat"),
                    "lon": event["client"].get("geographicalContext", {}).get("geolocation", {}).get("lon")
                },
                "result": "success" if event["outcome"]["result"] == "SUCCESS" else "failure",
                "failure_reason": event["outcome"].get("reason", ""),
                "app": event.get("target", [{}])[0].get("displayName", "Unknown"),
                "device": event["client"].get("device", "Unknown"),
                "browser": event["client"].get("userAgent", {}).get("browser", "Unknown"),
                "source": "okta"
            })
        elif log_source == "windows_ad":
            normalized.append({
                "timestamp": event["TimeCreated"],
                "user": event["TargetUserName"],
                "source_ip": event.get("IpAddress", ""),
                "location": None,  # Requires GeoIP enrichment
                "result": "success" if event["EventId"] in [4624, 4648] else "failure",
                "failure_reason": event.get("FailureReason", ""),
                "logon_type": event.get("LogonType", ""),
                "source": "windows_ad"
            })

    return pd.DataFrame(normalized)

# Enrich with GeoIP data for Windows AD logs missing location
import geoip2.database

def enrich_geoip(df, geoip_db_path="/opt/geoip/GeoLite2-City.mmdb"):
    """Add geolocation data to events missing location information."""
    reader = geoip2.database.Reader(geoip_db_path)

    for idx, row in df.iterrows():
        if row["location"] is None and row["source_ip"]:
            try:
                response = reader.city(row["source_ip"])
                df.at[idx, "location"] = {
                    "city": response.city.name,
                    "country": response.country.iso_code,
                    "lat": response.location.latitude,
                    "lon": response.location.longitude
                }
            except Exception:
                pass

    reader.close()
    return df

Step 2: Detect Impossible Travel Anomalies

Identify logins from geographically impossible locations:

from math import radians, sin, cos, sqrt, atan2
from datetime import datetime

def haversine_distance(lat1, lon1, lat2, lon2):
    """Calculate great-circle distance between two points in km."""
    R = 6371  # Earth's radius in kilometers

    lat1, lon1, lat2, lon2 = map(radians, [lat1, lon1, lat2, lon2])
    dlat = lat2 - lat1
    dlon = lon2 - lon1

    a = sin(dlat/2)**2 + cos(lat1) * cos(lat2) * sin(dlon/2)**2
    c = 2 * atan2(sqrt(a), sqrt(1-a))

    return R * c

def detect_impossible_travel(df, max_speed_kmh=900):
    """
    Detect impossible travel events where a user authenticates from
    two locations faster than physically possible.

    max_speed_kmh: Maximum realistic travel speed (900 km/h ~= commercial flight)
    """
    alerts = []

    # Sort by user and timestamp
    df_sorted = df.sort_values(["user", "timestamp"])

    for user, user_events in df_sorted.groupby("user"):
        successful_events = user_events[user_events["result"] == "success"]

        for i in range(1, len(successful_events)):
            prev = successful_events.iloc[i-1]
            curr = successful_events.iloc[i]

            # Skip if location data is missing
            if not prev.get("location") or not curr.get("location"):
                continue
            if not prev["location"].get("lat") or not curr["location"].get("lat"):
                continue

            # Calculate distance and time delta
            distance_km = haversine_distance(
                prev["location"]["lat"], prev["location"]["lon"],
                curr["location"]["lat"], curr["location"]["lon"]
            )

            time_diff = (pd.Timestamp(curr["timestamp"]) -
                        pd.Timestamp(prev["timestamp"])).total_seconds() / 3600

            if time_diff <= 0:
                continue

            required_speed = distance_km / time_diff

            # Flag if required speed exceeds maximum realistic travel
            if required_speed > max_speed_kmh and distance_km > 100:
                alerts.append({
                    "alert_type": "IMPOSSIBLE_TRAVEL",
                    "severity": "HIGH",
                    "user": user,
                    "timestamp": curr["timestamp"],
                    "details": {
                        "location_1": f"{prev['location']['city']}, {prev['location']['country']}",
                        "location_2": f"{curr['location']['city']}, {curr['location']['country']}",
                        "time_1": prev["timestamp"],
                        "time_2": curr["timestamp"],
                        "distance_km": round(distance_km, 1),
                        "time_hours": round(time_diff, 2),
                        "required_speed_kmh": round(required_speed, 1),
                        "source_ip_1": prev["source_ip"],
                        "source_ip_2": curr["source_ip"]
                    }
                })

    return alerts

# Run impossible travel detection
travel_alerts = detect_impossible_travel(auth_df)
print(f"Impossible travel alerts: {len(travel_alerts)}")
for alert in travel_alerts:
    print(f"  [{alert['severity']}] {alert['user']}: "
          f"{alert['details']['location_1']} -> {alert['details']['location_2']} "
          f"({alert['details']['distance_km']} km in {alert['details']['time_hours']}h)")

Step 3: Detect Brute Force and Password Spraying

Identify credential attack patterns across authentication logs:

from collections import Counter

def detect_brute_force(df, threshold_failures=10, window_minutes=10):
    """
    Detect brute force attacks: many failed attempts against
    a single account in a short time window.
    """
    alerts = []
    failed = df[df["result"] == "failure"].copy()
    failed["timestamp"] = pd.to_datetime(failed["timestamp"])

    for user, user_fails in failed.groupby("user"):
        user_fails_sorted = user_fails.sort_values("timestamp")

        # Sliding window analysis
        for i, row in user_fails_sorted.iterrows():
            window_start = row["timestamp"]
            window_end = window_start + timedelta(minutes=window_minutes)

            window_events = user_fails_sorted[
                (user_fails_sorted["timestamp"] >= window_start) &
                (user_fails_sorted["timestamp"] <= window_end)
            ]

            if len(window_events) >= threshold_failures:
                source_ips = window_events["source_ip"].unique()
                alerts.append({
                    "alert_type": "BRUTE_FORCE",
                    "severity": "HIGH",
                    "user": user,
                    "timestamp": str(window_start),
                    "details": {
                        "failed_attempts": len(window_events),
                        "window_minutes": window_minutes,
                        "source_ips": list(source_ips),
                        "distributed": len(source_ips) > 1,
                        "failure_reasons": dict(Counter(window_events["failure_reason"]))
                    }
                })
                break  # One alert per user per detection pass

    return alerts

def detect_password_spray(df, threshold_users=10, window_minutes=30):
    """
    Detect password spraying: failed logins against many different
    accounts from the same source in a short window (1-2 attempts per user).
    """
    alerts = []
    failed = df[df["result"] == "failure"].copy()
    failed["timestamp"] = pd.to_datetime(failed["timestamp"])

    for source_ip, ip_events in failed.groupby("source_ip"):
        ip_events_sorted = ip_events.sort_values("timestamp")

        for i, row in ip_events_sorted.iterrows():
            window_start = row["timestamp"]
            window_end = window_start + timedelta(minutes=window_minutes)

            window_events = ip_events_sorted[
                (ip_events_sorted["timestamp"] >= window_start) &
                (ip_events_sorted["timestamp"] <= window_end)
            ]

            unique_users = window_events["user"].nunique()
            attempts_per_user = len(window_events) / unique_users if unique_users > 0 else 0

            # Password spray: many users targeted, few attempts per user
            if unique_users >= threshold_users and attempts_per_user <= 3:
                # Check if any succeeded (compromised account)
                success_after = df[
                    (df["source_ip"] == source_ip) &
                    (df["result"] == "success") &
                    (pd.to_datetime(df["timestamp"]) > window_start) &
                    (pd.to_datetime(df["timestamp"]) < window_end + timedelta(hours=1))
                ]

                alerts.append({
                    "alert_type": "PASSWORD_SPRAY",
                    "severity": "CRITICAL" if len(success_after) > 0 else "HIGH",
                    "timestamp": str(window_start),
                    "details": {
                        "source_ip": source_ip,
                        "targeted_users": unique_users,
                        "total_attempts": len(window_events),
                        "avg_attempts_per_user": round(attempts_per_user, 1),
                        "window_minutes": window_minutes,
                        "successful_logins_after": len(success_after),
                        "compromised_accounts": list(success_after["user"].unique()) if len(success_after) > 0 else []
                    }
                })
                break

    return alerts

# Run detections
brute_force_alerts = detect_brute_force(auth_df)
spray_alerts = detect_password_spray(auth_df)
print(f"Brute force alerts: {len(brute_force_alerts)}")
print(f"Password spray alerts: {len(spray_alerts)}")

Step 4: Build Behavioral Baselines and Detect Deviations

Create user behavioral profiles and flag statistical anomalies:

import numpy as np
from scipy import stats
from sklearn.ensemble import IsolationForest

def build_user_baseline(df, user, lookback_days=90):
    """Build behavioral baseline for a specific user."""
    user_events = df[df["user"] == user].copy()
    user_events["timestamp"] = pd.to_datetime(user_events["timestamp"])
    user_events["hour"] = user_events["timestamp"].dt.hour
    user_events["day_of_week"] = user_events["timestamp"].dt.dayofweek

    baseline = {
        "user": user,
        "typical_hours": {
            "start": int(user_events["hour"].quantile(0.05)),
            "end": int(user_events["hour"].quantile(0.95)),
            "mean": float(user_events["hour"].mean()),
            "std": float(user_events["hour"].std())
        },
        "typical_days": list(user_events["day_of_week"].mode().values),
        "typical_ips": list(user_events["source_ip"].value_counts().head(10).index),
        "typical_locations": list(
            user_events["location"].apply(
                lambda x: x.get("country") if isinstance(x, dict) else None
            ).dropna().value_counts().head(5).index
        ),
        "typical_apps": list(user_events["app"].value_counts().head(10).index),
        "typical_devices": list(user_events["device"].value_counts().head(5).index),
        "avg_daily_logins": float(
            user_events.groupby(user_events["timestamp"].dt.date).size().mean()
        ),
        "std_daily_logins": float(
            user_events.groupby(user_events["timestamp"].dt.date).size().std()
        ),
        "failure_rate": float(
            (user_events["result"] == "failure").mean()
        )
    }

    return baseline

def detect_behavioral_anomalies(event, baseline):
    """Compare a new authentication event against user baseline."""
    anomalies = []
    event_time = pd.Timestamp(event["timestamp"])

    # Off-hours login detection
    hour = event_time.hour
    if baseline["typical_hours"]["std"] > 0:
        z_score = abs(hour - baseline["typical_hours"]["mean"]) / baseline["typical_hours"]["std"]
        if z_score > 2.5:
            anomalies.append({
                "type": "OFF_HOURS_LOGIN",
                "severity": "MEDIUM",
                "detail": f"Login at {hour}:00 (baseline: {baseline['typical_hours']['start']}:00-{baseline['typical_hours']['end']}:00)",
                "z_score": round(z_score, 2)
            })

    # New source IP
    if event["source_ip"] not in baseline["typical_ips"]:
        anomalies.append({
            "type": "NEW_SOURCE_IP",
            "severity": "MEDIUM",
            "detail": f"Login from unknown IP: {event['source_ip']}"
        })

    # New country
    if event.get("location") and isinstance(event["location"], dict):
        country = event["location"].get("country")
        if country and country not in baseline["typical_locations"]:
            anomalies.append({
                "type": "NEW_COUNTRY",
                "severity": "HIGH",
                "detail": f"Login from new country: {country}"
            })

    # New application
    if event.get("app") and event["app"] not in baseline["typical_apps"]:
        anomalies.append({
            "type": "NEW_APPLICATION",
            "severity": "LOW",
            "detail": f"Access to new application: {event['app']}"
        })

    # New device
    if event.get("device") and event["device"] not in baseline["typical_devices"]:
        anomalies.append({
            "type": "NEW_DEVICE",
            "severity": "MEDIUM",
            "detail": f"Login from new device: {event['device']}"
        })

    # Weekend login for weekday-only users
    if event_time.dayofweek >= 5 and 5 not in baseline["typical_days"] and 6 not in baseline["typical_days"]:
        anomalies.append({
            "type": "WEEKEND_LOGIN",
            "severity": "LOW",
            "detail": f"Weekend login detected (typical days: {baseline['typical_days']})"
        })

    return anomalies

def isolation_forest_anomaly_detection(df):
    """Use Isolation Forest for multivariate anomaly detection."""
    # Feature engineering
    features_df = df.copy()
    features_df["timestamp"] = pd.to_datetime(features_df["timestamp"])
    features_df["hour"] = features_df["timestamp"].dt.hour
    features_df["day_of_week"] = features_df["timestamp"].dt.dayofweek
    features_df["is_failure"] = (features_df["result"] == "failure").astype(int)

    # Encode categorical features
    features_df["ip_frequency"] = features_df.groupby("source_ip")["source_ip"].transform("count")
    features_df["user_frequency"] = features_df.groupby("user")["user"].transform("count")

    feature_columns = ["hour", "day_of_week", "is_failure", "ip_frequency", "user_frequency"]
    X = features_df[feature_columns].fillna(0)

    # Train Isolation Forest
    model = IsolationForest(
        n_estimators=200,
        contamination=0.01,  # Expect 1% anomaly rate
        random_state=42,
        n_jobs=-1
    )
    features_df["anomaly_score"] = model.fit_predict(X)
    features_df["anomaly_probability"] = model.score_samples(X)

    # Extract anomalies (labeled as -1)
    anomalies = features_df[features_df["anomaly_score"] == -1]

    return anomalies.sort_values("anomaly_probability")

Step 5: Implement SIEM Detection Rules

Deploy detection rules for common authentication attack patterns:

# Splunk SPL queries for authentication anomaly detection

# 1. Brute Force Detection
# name: Authentication Brute Force - Multiple Failed Logins
# severity: high
brute_force_spl: |
  index=auth sourcetype IN ("azure:aad:signin", "okta:im:log", "WinEventLog:Security")
  (result="failure" OR EventCode=4625)
  | bin _time span=10m
  | stats count as failed_attempts dc(src_ip) as unique_ips
    values(src_ip) as source_ips
    latest(_time) as last_attempt
    by user _time
  | where failed_attempts >= 10
  | eval alert_type=if(unique_ips > 3, "Distributed Brute Force", "Standard Brute Force")

# 2. Password Spray Detection
# name: Password Spray Attack - Multiple Users Same Source
# severity: critical
password_spray_spl: |
  index=auth sourcetype IN ("azure:aad:signin", "okta:im:log")
  result="failure"
  | bin _time span=30m
  | stats dc(user) as targeted_users count as total_attempts
    values(user) as users_targeted
    by src_ip _time
  | where targeted_users >= 10
  | eval attempts_per_user = round(total_attempts / targeted_users, 1)
  | where attempts_per_user <= 3
  | eval severity=if(targeted_users > 50, "CRITICAL", "HIGH")

# 3. Impossible Travel Detection
# name: Impossible Travel - Geographically Inconsistent Logins
# severity: high
impossible_travel_spl: |
  index=auth result="success"
  | iplocation src_ip
  | sort user _time
  | streamstats current=f last(lat) as prev_lat last(lon) as prev_lon
    last(_time) as prev_time last(City) as prev_city last(Country) as prev_country
    by user
  | where isnotnull(prev_lat) AND isnotnull(lat)
  | eval distance_km = 6371 * 2 * asin(sqrt(
      pow(sin((lat - prev_lat) * pi() / 360), 2) +
      cos(prev_lat * pi() / 180) * cos(lat * pi() / 180) *
      pow(sin((lon - prev_lon) * pi() / 360), 2)))
  | eval time_hours = (_time - prev_time) / 3600
  | eval required_speed = distance_km / time_hours
  | where required_speed > 900 AND distance_km > 100

# 4. Credential Stuffing Detection
# name: Credential Stuffing - High Volume Failed Logins with Some Successes
# severity: critical
credential_stuffing_spl: |
  index=auth
  | bin _time span=1h
  | stats count(eval(result="failure")) as failures
    count(eval(result="success")) as successes
    dc(user) as unique_users
    dc(src_ip) as unique_ips
    by src_ip _time
  | where failures > 100 AND successes > 0 AND unique_users > 20
  | eval success_rate = round(successes / (failures + successes) * 100, 2)
  | where success_rate < 5

Step 6: Correlate and Score Authentication Anomalies

Combine multiple detection signals into risk scores:

def calculate_auth_risk_score(user, alerts, baseline):
    """
    Calculate composite risk score for authentication events.
    Combines multiple anomaly signals with weighted scoring.
    """
    score = 0
    risk_factors = []

    weights = {
        "IMPOSSIBLE_TRAVEL": 40,
        "PASSWORD_SPRAY": 35,
        "BRUTE_FORCE": 30,
        "CREDENTIAL_STUFFING": 35,
        "NEW_COUNTRY": 25,
        "OFF_HOURS_LOGIN": 15,
        "NEW_SOURCE_IP": 10,
        "NEW_DEVICE": 10,
        "NEW_APPLICATION": 5,
        "WEEKEND_LOGIN": 5,
        "MFA_BYPASS": 45,
        "LEGACY_PROTOCOL": 20
    }

    for alert in alerts:
        alert_type = alert.get("type") or alert.get("alert_type")
        weight = weights.get(alert_type, 10)

        # Adjust weight based on severity
        severity_multiplier = {
            "CRITICAL": 2.0,
            "HIGH": 1.5,
            "MEDIUM": 1.0,
            "LOW": 0.5
        }
        severity = alert.get("severity", "MEDIUM")
        adjusted_weight = weight * severity_multiplier.get(severity, 1.0)

        score += adjusted_weight
        risk_factors.append({
            "factor": alert_type,
            "weight": adjusted_weight,
            "detail": alert.get("detail", alert.get("details", ""))
        })

    # Normalize score to 0-100
    normalized_score = min(100, score)

    # Determine risk level
    if normalized_score >= 80:
        risk_level = "CRITICAL"
        recommended_action = "Immediate account suspension and investigation"
    elif normalized_score >= 60:
        risk_level = "HIGH"
        recommended_action = "Force MFA re-enrollment and notify SOC"
    elif normalized_score >= 40:
        risk_level = "MEDIUM"
        recommended_action = "Require step-up authentication"
    elif normalized_score >= 20:
        risk_level = "LOW"
        recommended_action = "Monitor and log for trend analysis"
    else:
        risk_level = "INFORMATIONAL"
        recommended_action = "No action required"

    return {
        "user": user,
        "risk_score": normalized_score,
        "risk_level": risk_level,
        "recommended_action": recommended_action,
        "risk_factors": sorted(risk_factors, key=lambda x: x["weight"], reverse=True),
        "timestamp": datetime.utcnow().isoformat()
    }

Key Concepts

Tools & Systems

• Microsoft Sentinel UEBA: Cloud-native SIEM with built-in entity behavior analytics for Azure AD and multi-cloud authentication anomaly detection
• Exabeam Advanced Analytics: UEBA platform using machine learning for user session analysis and automated threat timeline construction
• Splunk UBA: Behavioral analytics add-on for Splunk providing pre-built authentication anomaly models and risk scoring
• Elastic SIEM ML Jobs: Machine learning anomaly detection jobs for authentication log analysis in the Elastic Stack

Common Scenarios

Scenario: Detecting Compromised Executive Account After Password Spray

Context: SOC observes a spike in failed authentication attempts from a cloud VPS IP address targeting 200+ accounts. Two hours later, an executive account shows successful authentication from the same IP range followed by mailbox rule creation and data exfiltration.

Approach:

1. Run password spray detection across the timeframe to identify all targeted accounts
2. Cross-reference targeted accounts with subsequent successful logins from related IP ranges
3. Build behavioral baseline for the executive account and flag all deviations
4. Check for impossible travel between the executive's last legitimate login and the attacker's session
5. Identify post-compromise activity: mailbox rules, file downloads, delegated access changes
6. Calculate composite risk score combining password spray, new IP, off-hours login, and new device signals
7. Trigger automated response: force session termination, disable account, notify manager

Pitfalls:

• Relying on single-signal detection (failed logins only) misses successful spray results
• Not correlating across identity providers when users have accounts in multiple IdPs
• Static thresholds that do not account for legitimate VPN IP changes or travel
• Ignoring successful authentications after the spray window closes (attackers may wait before using credentials)

Output Format

AUTHENTICATION ANOMALY DETECTION REPORT
=========================================
Analysis Period:   2026-02-01 to 2026-02-24
Total Auth Events: 2,847,392
Users Monitored:   3,847
Alert Sources:     Azure AD, Okta, Windows AD

THREAT DETECTION SUMMARY
Password Spray Attacks:    3
Brute Force Attacks:       12
Impossible Travel:         8
Credential Stuffing:       1
Behavioral Anomalies:      47

HIGH-RISK ACCOUNTS
[CRITICAL] j.smith@corp.com     Score: 92
  - Impossible travel: Chicago -> Moscow (7,876 km in 0.5h)
  - Password spray target followed by successful login
  - New device and browser fingerprint
  - Off-hours access to SharePoint and email
  Action: Account suspended, SOC investigation initiated

[HIGH] m.johnson@corp.com       Score: 67
  - Login from new country (Brazil)
  - New source IP not matching VPN ranges
  - Access to HR application outside normal pattern
  Action: MFA re-enrollment required, manager notified

[MEDIUM] a.williams@corp.com    Score: 38
  - Weekend login at 03:00 UTC
  - New device (Linux, typically Windows user)
  Action: Step-up authentication applied

ATTACK CAMPAIGN DETAILS
Password Spray Campaign #1:
  Source:            185.220.101.x/24 (Tor exit node)
  Targeted Users:    247
  Success Rate:      0.8% (2 accounts compromised)
  Compromised:       j.smith@corp.com, r.davis@corp.com
  Duration:          45 minutes
  Pattern:           2 attempts per user, 3-second interval