Inessa Costa Duc
Back to Projects
Machine Learning

Predictive Analytics: Earthquake Damage

Predicting building damage severity to support disaster response

🔍 The Problem

Following the 2015 Gorkha earthquake in Nepal, emergency response teams needed a way to rapidly assess which buildings were most likely to suffer severe damage, enabling faster allocation of limited relief resources.

🧭 Approach

Worked with a dataset of 260,000+ buildings, engineering features from structural properties, geographical data, and construction materials. Designed and evaluated multiple ML models from logistic regression to Neural Networks and XGBoost.

⚙️ Solution

Built a multi-class classification model achieving top-tier accuracy on the DrivenData competition benchmark. Leveraged Cursor AI to accelerate the development pipeline.

📈 Impact

Drove a 14% improvement in predictive performance. Demonstrated how predictive analytics can accelerate humanitarian response by prioritising high-risk structures.

Tools & Technologies

PythonXGBoostNeural NetworksScikit-learnCursor AI