Scribd
Upload
21 views10 pages

Hackathon

The document presents a machine learning model aimed at predicting road accident severity to enhance safety and emergency response. It discusses the challenges faced, the methodology employed, and the key insights derived from the analysis of historical accident data. Recommendations for road safety improvements and future work directions are also outlined.

Uploaded by

Ojaswi Sonawane
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
21 views10 pages

Hackathon

The document presents a machine learning model aimed at predicting road accident severity to enhance safety and emergency response. It discusses the challenges faced, the methodology employed, and the key insights derived from the analysis of historical accident data. Recommendations for road safety improvements and future work directions are also outlined.

Uploaded by

Ojaswi Sonawane
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

Predicting Road Accident Severity Using

Machine Learning

Presented by: Code Commandos


• Ojaswi Sonawane (Team Leader)
• Arya Pandey
• Dipti Kadam
• Jagruti Sakore
• Gauri Suradkar
Introduction
• Problem Statement:
– Road accidents are a growing public safety concern with
varying degrees of severity (minor injuries to fatal crashes).
– Accurate predictions of crash severity can enhance road
safety and emergency response strategies.
• Objective:
– Build a machine learning model to predict crash severity.
– Analyze key risk factors (e.g., road design, weather, driver
behavior).
– Offer actionable road safety recommendations.
• Dataset Overview:
– Historical accident data with
features like speed, vehicle
type, road conditions, and
more.
– Crash severity levels: Minor

Problem Injury, Major Injury, Fatal


Crash.
• Challenges:

Analysis – Balancing the dataset to


avoid bias.
– Selecting meaningful features
for accurate predictions.
– Ensuring the model
generalizes well.
• Methodology:
– Data preprocessing
(handling categorical and
numerical data, missing
values).
– Feature engineering to
identify key contributors to
Proposed crash severity.
– Using a Gradient Boosting
Classifier for predictions.
Solution • Hyperparameter Optimization:
– Performed with
GridSearchCV for
parameters like
n_estimators,
learning_rate, and
max_depth.
Model Performance
• Accuracy:
• Achieved {accuracy}% accuracy on the test set.
• Confusion Matrix:
• Visualized using a heatmap.
• Highlights predictions for Minor, Major, and Fatal
crashes.
• Classification Report:
• Precision, Recall, and F1-scores for each severity level.
Key-Insights
Feature importance:
• Highlight the top 3-5 factors contributing to severe crashes
(e.g., speed, alcohol consumption, road conditions).

Risk patterns:
• Higher severity in wet/icy conditions.
• Influence of road type and vehicle type on crash outcomes.
Recommendations • Road Safety Improvements:
– Implement stricter speed
limits in high-risk zones.
– Enhance road surface
maintenance and signage.
– Increase awareness programs
for seatbelt usage and alcohol
consumption risks.
• Policy Interventions:
– Monitor and optimize urban
and rural road designs.
– Incorporate real-time
weather and traffic
monitoring systems.
• Short-Term:
– Better emergency response
through severity prediction.
• Long-Term:
– Reduction in accidents and
fatalities by identifying and Impact
mitigating risk factors.
– Data-driven policymaking for
road safety enhancements.
Conclusion

• Summary:
• Successfully built a model achieving the required accuracy.
• Identified key crash severity factors and proposed actionable
recommendations.
• Future Work:
• Integrate additional data sources (e.g., real-time traffic data).
• Explore advanced models like deep learning for improved
accuracy.
Thank you for your
attention!

You might also like