🛡️ ATLShield: Crime Forecasting and Patrol Recommendation System

ATLShield is a Flask-based API that uses machine learning models to:

• 🔍 Predict hourly crime volumes for a given neighborhood and month
• 🚓 Recommend the best patrolling strategy per hour
• 📊 Generate visualizations to help law enforcement with decision-making

All predictions are backed by trained Random Forest models using real Atlanta crime data.

---

⚙️ Setup Instructions

Python backend: Run on AWS! -- t2.small instance

1. Clone the repo and enter the directory

   git clone <repo-url>
   cd atlShield
   

2. Give access to Shell script

   chmod +x run.sh
   

3. Run Shell script to start Python service with nohup

   ./run.sh
   

4. Tail logs

   tail -f atlShield.log
   

=======

📊 Tableau Visualizations

All dashboards developed as part of this project are published on Tableau Public.

You can also download the workbook from the same link to explore the design, layout, and implementation details.

---

📁 Project Structure

Exploratory Data Analysis (EDA)

Before building the prediction models, we conducted a thorough EDA on Atlanta crime data from 2021 to 2025. Key steps:

🔍 Data Cleaning

• Missing Value Handling:

  • Dropped columns with more than 30% missing values.
  • Rows missing critical fields (e.g., Zone, Beat, LocationType, ReportDate, Latitude, Longitude) were removed to maintain data integrity.

• Date Fixes:

  • Cleaned and standardized ReportDate.
  • If ReportDate was invalid or missing, it was imputed using OccurredToDate.

Offense Mapping

• Categorized offenses into Group A and Group B using FBI’s NIBRS offense codes.
• Created an offense_code_map dictionary to easily filter and group crime types.

Why this matters?

These preprocessing steps ensured that:

• The model was trained on high-quality, consistent data.
• Time-based trends (like seasonal crime variations or hourly patterns) were accurately captured.
• Crime types could later be analyzed for patrolling strategy recommendations and clustering (hotspot analysis).

---

Code Overview

• atlShield.py hosts a Flask backend with two endpoints:
  • /predict_hourly_crime: Predicts hourly crime volume (using a RandomForestRegressor) and recommends patrol strategies (using a RandomForestClassifier) based on historical data, time, and location.
  • /average_daily_crime: Returns the average number of crimes per day for a given neighborhood and month.
• Models are trained at startup using grouped, encoded features from updated_patrol_types.csv.
• Visual output: A crime-intensity bar chart is generated (with red/yellow/green bars) per request and saved to disk.
• Everything runs on a headless backend (matplotlib.use('Agg')) to ensure compatibility with servers like AWS EC2.
• Hosted on AWS EC2 instance for scalable deployment.
• Integrated with Tableau (via locally running TabPy) for interactive, real-time predictions directly from dashboards.