Traffic Violation Predictions

Traffic Violation Predictions is a machine learning web app built to analyze and predict various traffic-related risks using over 1.6 million traffic stop records from Montgomery County, Maryland (2012 to 2025). The application provides insights into high-risk periods, accident likelihood, and potential violations by location. It is designed to support data-driven decisions in public safety and enforcement planning.

Project Overview

This project uses a large-scale dataset to explore and model three main traffic safety tasks:

• High-Risk Time Prediction by Location
• Accident Occurrence Prediction
• Violation Probability Estimation by Coordinates
• Each task is powered by a custom-trained machine learning model and integrated into a lightweight web application for easy user interaction.

Dataset Details

• Source: Data.gov
• Region: Montgomery County, Maryland
• Date Range: 2012 – 2025
• Size: ~1.6 million records (~809 MB CSV)

Key Features:

• Stop time and location (latitude, longitude)
• Violation type
• Driver and vehicle details
• Accident, injury, and damage indicators

Data Preprocessing Highlights

• Standardized column names for readability and consistency
• Extracted temporal features (hour, day of week, month)
• Cleaned and validated spatial coordinates
• Converted binary indicators to Boolean types
• Handled missing values via imputation or row removal
• Encoded categorical features and applied feature scaling
• Removed duplicate records to prevent training bias

Machine Learning Models

1. High-Risk Period Prediction by Location

Model: Random Forest Classifier

Objective: Predict whether a stop occurs during a high-risk time block based on spatial and temporal features.

• Stop locations clustered into 50 zones using K-Means
• Labeled time blocks as "high-risk" if in top 25% of violation frequency
• Features: location cluster, hour, day of week, month
• Output: Top 10 high-risk zones ranked by predicted probability

2. Accident Prediction Model

Model: Random Forest Classifier (with SMOTE for class balancing)

Objective: Predict whether a traffic stop results in an accident.

• Features include time, driver demographics, vehicle type, and violation context
• Applied label encoding, one-hot encoding, and standardization
• Balanced class distribution using SMOTE to handle rare accident cases
• Output: Probability of accident for each stop record

3. Violation Probability Estimator by Location

Model: Gradient Boosting Classifier

Objective: Predict the probability of specific violation types based solely on location (latitude & longitude).

• One Gradient Boosting Classifier model per violation type: accident, alcohol, personal injury, property damage, fatal, hazmat, seat belt
• Returns top 4 most likely violations for any coordinate input
• Enables geospatial risk assessment and planning

How to Run the App

This is a Flask-based web app. To launch locally:

Step 1: Clone the repository

• git clone

Step 2: Install dependencies

• pip install -r requirements.txt

Step 3: Run the app

• python app.py
• Then open http://localhost:5000 in your browser.

The link to presentation : https://docs.google.com/presentation/d/1sgDy43Iur2q5BfV2G8ye_bhN6uijMA_2nr7FazKcdDk/edit?usp=sharing