AI Insights

AI-Powered Safe Route Prediction for Earthquakes

November 14, 2023

article featured image

In this article, we will explore how Omdena developed an AI-powered system to predict safe routes post-earthquake, utilizing machine learning algorithms and data analysis for immediate damage assessment, risk scoring, and route calculation to enhance public safety.


Earthquakes are among the most devastating natural disasters, causing widespread damage and loss of life. They pose significant challenges in assessing the safety of infrastructure post-disaster. The primary challenges include:

  • Immediate Damage Assessment: Quickly determining which buildings and roads remain intact and safe for use is a daunting task, further complicated by aftershocks and damaged utilities.
  • Safe Route Identification: In the chaos following an earthquake, identifying paths that are both safe and navigable becomes critical. The usual routes may be obstructed or dangerous, complicating efforts to reach safety, loved ones, or essential services.
  • Access to Emergency Services: The disruption to transportation networks can severely hamper the delivery of emergency services, including medical aid, to those in need.
  • Communication Barriers: With infrastructure possibly compromised, getting accurate information about safe areas and routes to the general public is a challenge.

These challenges underscore the necessity for a swift, reliable method to assess post-earthquake conditions and guide residents and responders safely through affected areas.


Omdena developed an AI-powered system that can predict safe routes in the aftermath of an earthquake. The system uses machine learning algorithms to analyze data on buildings, streets, and historical earthquake data to calculate a risk score for each section of the district. The system then uses this information to calculate the shortest and safest path between two points.

The models and processes used in developing this solution includes:

Data Collection and Preprocessing

The first step involves collecting a vast amount of data from multiple sources, including:

  • Structural Data: Information on buildings, such as construction material, age, height, and previous damage history.
  • Geographical Data: Detailed maps outlining streets, lanes, and pathways.
  • Historical Earthquake Data: Records of past earthquakes, including magnitude, depth, and affected areas.
  • Real-Time Sensor Data: Inputs from seismic activity monitors and other relevant sensors.

This data undergoes preprocessing to clean, normalize, and structure it for analysis. Missing values are addressed, and data is segmented according to geographical regions for localized processing.

Machine Learning Models

The system employs several ML models to analyze the preprocessed data:

  • Convolutional Neural Networks (CNNs): Used for analyzing satellite and aerial imagery to detect cracks and damages in buildings and roads that are not visible at the street level.
  • Recurrent Neural Networks (RNNs): Particularly Long Short-Term Memory (LSTM) networks, for predicting future seismic activity by learning from historical earthquake sequences.
  • Decision Trees: For classifying buildings and streets based on their risk level by considering various factors such as construction material, age, and proximity to fault lines.
  • Graph Neural Networks (GNNs): To model the city’s infrastructure as a graph where nodes represent key points (e.g., buildings, intersections) and edges represent paths (e.g., streets). GNNs help in understanding the connectivity and flow within the city to identify alternative routes.

Risk Assessment Process

  • Initial Assessment: Using CNNs to analyze imagery data for immediate damage assessment post-earthquake.
  • Risk Scoring: Each building and street segment is assigned a risk score based on factors like structural integrity, historical data, and predicted seismic activity. This uses a combination of Decision Trees and LSTM networks.
  • Route Calculation: GNNs calculate the safest and shortest path between two points considering the risk scores of all nodes (buildings, intersections) and edges (streets) in the path.

Continuous Learning

The system is designed to learn continuously by incorporating new data from sensors and user feedback. This ensures that the risk assessment models are always updated with the latest information, improving accuracy over time.

Simulation and Validation

Before deployment, the system is tested through simulations of various earthquake scenarios to validate its predictions against historical data and expert evaluations. This step is crucial for refining the models and ensuring their reliability in real-world situations.


The AI-powered safe route prediction system developed by Omdena has demonstrated remarkable effectiveness in ensuring public safety following earthquakes. In a comprehensive simulation of a magnitude 7.0 earthquake affecting Istanbul, the system’s performance was thoroughly evaluated. Here are the expanded results based on this simulation:

  • Accuracy in Route Prediction: The system achieved a high accuracy rate, successfully predicting safe routes for over 90% of Istanbul’s residents. This indicates the system’s reliability in identifying paths that avoid damaged infrastructure and areas at risk of aftershocks.
  • Coverage and Scalability: Despite Istanbul’s complex urban layout, which includes densely populated areas, narrow streets, and diverse building structures, the system managed to provide comprehensive coverage across the city. This suggests the system’s scalability and its potential applicability to other urban areas with varying geographical and architectural characteristics.
  • Response Time: The system demonstrated rapid processing capabilities, quickly generating safe route suggestions. This is critical in post-earthquake scenarios where time is of the essence for evacuation, rescue operations, and reaching safe zones.
  • User Feedback: Initial user feedback collected during the simulation emphasized the system’s ease of use and the clarity of the route instructions provided. This user-friendly interface is crucial for widespread adoption and utilization in crisis situations.
  • Comparison with Traditional Methods: When compared to traditional methods of route planning post-disaster, which often rely on manual assessment and can be time-consuming and less precise, Omdena’s AI system offered a significantly more efficient solution. By automating data analysis and leveraging real-time information, the system ensured more accurate and timely updates on safe passages.
  • Flexibility and Adaptability: The system displayed a high degree of flexibility, adapting to new data inputs such as reports of additional damage or blockages. This adaptability is essential for dealing with the dynamic and evolving nature of post-earthquake environments.


The AI-powered safe route prediction system has a number of potential benefits, including:

Reduced risk of injury and death: By providing users with safe routes to travel after an earthquake, the system can help to reduce the risk of injury and death.

Faster and safer family reunification: The system can help people to reunite with their loved ones more quickly and safely by providing them with information on the safest way to reach each other.

Improved emergency response: The system can help to improve emergency response by providing first responders with information on the safest way to reach affected areas.

Reduced costs: The system can help to reduce the costs of rebuilding and recovery after an earthquake by helping to prevent damage to property and infrastructure.

Successful Project between Omdena and ImpactHub Istanbul

In collaboration with ImpactHub Istanbul, Omdena completed a project to develop an AI prototype for predicting the safest routes during an earthquake in Istanbul. The project aimed to reunite families and mitigate disaster by using AI to identify safe paths to schools, hospitals, workplaces, and homes in the event of an earthquake. The team utilized various AI techniques to assess the safety of routes in Istanbul’s Fatih District and created a proof-of-concept for a deployable application to be used in earthquake response.

Figure 3: Path comparison — shortest and safest

Path comparison — shortest and safest. Source: Omdena

Find more information about this project here!


The AI-powered safe route prediction system developed by Omdena is a promising new technology that has the potential to save lives and improve emergency response in the aftermath of an earthquake. The system is currently being piloted in Istanbul, Turkey, and Omdena is working to deploy the system in other cities around the world.

Related case study: AI-Assisted Mapping Tool for Disaster Management

Want to work with us?

If you want to discuss a project or workshop, schedule a demo call with us by visiting:

media card
Leading Change in Crisis Management: Our DIMA Chatbot Success Story in Collaboration with DataCamp
media card
Top 62 Road Safety Companies and Organizations in 2024
media card
Advanced EDA of UK’s Road Safety Data using Python
media card
Devising a Safe Path To Travel In The Aftermath Of An Earthquake