This project presents a deep learning-based framework for identifying optimal ambulance positioning in urban areas to improve emergency response times. By analyzing real-time accident, road segment, and weather data from Nairobi, it leverages Deep Embedded Clustering (DEC) and Cat2Vec embeddings to predict accident hotspots and suggest strategic ambulance deployment locations.
- 📊 Real-time analysis of traffic accidents, road conditions, and weather data
- 🧠 Deep Embedded Clustering (DEC) for precise hotspot detection
- 🧬 Cat2Vec: Deep embeddings for high-cardinality categorical variables
- 🗺️ Predicts coordinates for optimal ambulance positions in a city
- 📈 Achieves 95% clustering accuracy using K-Fold cross-validation
- ✅ Outperforms traditional clustering algorithms (K-Means, GMM, Agglomerative)
- Deep Embedded Clustering (DEC) with autoencoders
- Cat2Vec: Neural network embeddings for categorical data
- Exploratory Data Analysis (EDA) to extract accident patterns
- Custom distance scoring for real-world validation of cluster centroids
- Evaluation Metrics: Silhouette Score, Calinski-Harabasz Index, Davies-Bouldin Index, and a novel distance-based metric
This project is based on my published research. 📄 Read the full paper here.