CRIMEX v1 — LA Incident Behavioral Intelligence Dataset

📌 Overview

CRIMEX v1 is a large-scale enriched criminal incident dataset derived from the Los Angeles Open Crime Data.

It transforms raw incident records into a behavioral intelligence dataset by adding multiple feature layers:

• Temporal intelligence
• Behavioral profiling
• Contextual risk signals
• Graph-inspired features
• Data quality and explainability metrics

The dataset is designed for machine learning, crime analytics, and research applications.

---

📊 Dataset Summary

• Total records: ~1,004,894
• Total features: 150+
• Location: Los Angeles, USA
• Time span: Based on source data
• Format: Parquet (full), CSV (sample)

---

🧠 Feature Categories

1. Core Incident Data

• Case ID, crime type, timestamps
• Victim attributes
• Weapon and premise information

2. Temporal Features

• Hour, day, weekday, weekend
• Seasonal indicators
• Time-based patterns

3. Behavioral Intelligence

• Modus operandi complexity
• Multi-step behavior indicators
• Behavioral risk scoring
• Criminal sophistication proxy

4. Contextual Risk Signals

• Opportunity context
• Guardianship proxy
• Routine activity theory features
• Environmental stress indicators (proxy-based)

5. Graph-Inspired Features

• Behavior signature keys
• Connectivity approximations
• Centrality and exposure proxies

6. Explainability Layer

• Risk explanation text
• Interpretable feature contributions
• Risk categorization codes

7. Data Quality

• Completeness score
• Consistency flags
• Enrichment confidence

---

🌍 Geo Information

• Latitude/Longitude included
• Missing coordinates partially recovered via geocoding
• Geo/POI enrichment planned in future versions

---

🚧 Limitations

• Derived from a single city (Los Angeles)
• Some features are heuristic or proxy-based
• No full POI/environment enrichment yet (planned in v2)
• Actor-level linking not included in this version

---

🚀 Use Cases

• Crime prediction modeling
• Risk scoring systems
• Behavioral pattern analysis
• Graph-based crime analysis
• Anomaly detection
• Explainable AI research

---

⚖️ Data Source & License

This dataset is derived from:

• Los Angeles Open Data Portal
  https://data.lacity.org

Users must comply with the original data source license.

CRIMEX v1 provides feature-engineered transformations and does not modify original legal ownership.

---

CRIMEX v2 — Actor Intelligence Dataset

📌 Overview

CRIMEX v2 is a person-level intelligence dataset built from public watchlists and criminal data sources.

It extends the CRIMEX framework from incident-level analysis (v1) to actor-level intelligence modeling.

The dataset transforms raw OpenSanctions data into:

• Structured actor profiles
• Crime-aware intelligence features
• Risk scoring and prioritization
• Explainable investigation signals

---

📊 Dataset Summary

• Source: OpenSanctions Crime Dataset
• Raw records: 247,463
• Final actor records: 170,097 :contentReference[oaicite:1]{index=1}
• Total features: 112 :contentReference[oaicite:2]{index=2}
• Format: Parquet
• Type: Tabular intelligence dataset

---

🧠 Feature Categories

1. Identity Features

• Actor name and aliases
• Birth date and derived age
• Nationality and multi-country indicators
• Identity complexity score (aliases + identifiers)

---

2. Crime Intelligence

• Cleaned crime description

• Enhanced crime text extraction

• Crime ontology classification:

  • violent
  • fraud_identity
  • narcotics
  • child_related
  • other
  • unknown

• Serious crime flag

• Crime text quality indicators

---

3. Source Intelligence

• Source dataset name (e.g., INTERPOL, sanctions lists)
• Watchlist indicators
• Terrorism flag
• Source risk score

---

4. Temporal Features

• First seen / last seen dates
• Observation duration
• Days since last update
• Recent activity indicators

---

5. Geographic Context

• Country count
• Cross-border activity flag
• Address-based mobility score

---

6. Risk Scoring

CRIMEX introduces a multi-layer risk system:

• actor_risk_score_v3 → numerical score
• actor_risk_level_v3 → low / medium / high

Risk is derived from:

• crime type
• source reliability
• temporal activity
• identity complexity
• geographic spread

---

7. Investigator Intelligence

Designed for real-world usage:

• investigator_priority_level_refined
• investigator_priority_score
• investigator_segment

Segments include:

• urgent_high_risk
• high_risk_monitor
• manual_review
• medium_risk_watch
• low_risk_archive

---

8. Explainability Layer

Each actor includes human-readable explanations:

• risk_explanation

Examples:

• serious_crime|watchlist_source|cross_border
• multiple_identities|long_term_activity
• low_signal

---

9. Data Quality Features

• Data completeness score
• Missing field indicators
• Data confidence score
• Sparse data flags

---

🔬 Key Contributions

CRIMEX Actor Intelligence introduces:

• Multi-source actor intelligence modeling
• Explainable risk scoring
• Separation of risk vs investigator priority
• Behavioral + contextual + temporal fusion
• Analyst-ready feature design

---

🚀 Use Cases

• Criminal risk scoring
• Watchlist prioritization
• Law enforcement analytics
• Graph intelligence systems
• Entity resolution and linking
• Explainable AI research

---

⚠️ Limitations

• Based on public watchlists (coverage varies by country)
• Crime descriptions are partially inferred from source text
• Some ontology mapping is rule-based
• Not all actors have complete identity data

---

⚖️ Data Source & License

Derived from:

• OpenSanctions
  https://www.opensanctions.org

Important:

• Data is subject to OpenSanctions license (CC BY-NC 4.0)
• Users must comply with attribution and non-commercial use

CRIMEX provides feature-engineered transformations only.

---

Reproducibility

The dataset is fully reproducible using:

• Notebook pipeline
• Modular Python scripts

Pipeline modules:

• src/actor.py
• src/actor_pipeline.py
• src/actor_extended.py :contentReference[oaicite:3]{index=3}

---

Files Included

• crimex_actor_intelligence.parquet → final dataset
• crimex_actor_feature_dictionary.csv → feature documentation
• crimex_actor_dataset_metadata.json → dataset metadata

---

Future Work

• Multi-dataset actor linking
• Graph-based intelligence networks
• Real-time watchlist updates
• Integration with CRIMEX v1 (incident + actor fusion)

---

Future Versions

• CRIMEX v2: Geo/POI enrichment using OpenStreetMap (local stack)
• CRIMEX v3: Multi-source integration (OpenSanctions, Interpol, etc.)
• CRIMEX v4: Global crime intelligence graph

---

👤 Author

Ayman Sharara
Data Scientist | AI Researcher | Crime Intelligence Systems

---

📬 Feedback

Contributions and feedback are welcome.