Wildfire Spread Prediction
This project applies machine learning to predict wildfire risk in California, using real-world weather and vegetation data. I helped to designed and trained a neural network that analyzes over 237,000 wildfire occurrences to help improve resource planning and emergency response.
Key Skills Demonstrated
- Processed and cleaned large datasets using Python and Pandas to extract meaningful patterns from raw weather, vegetation, and wildfire occurrence data.
- Designed and trained neural network models using TensorFlow and PyTorch, reaching an 87% prediction accuracy for wildfire risk classification.
- Built and compared multiple models with scikit-learn to evaluate different algorithms and select optimal architectures based on performance and generalization.
- Visualized results and model behavior using Matplotlib to better understand predictions and guide further refinement.
- Managed data ingestion, feature engineering, and hyperparameter tuning to balance model complexity with accuracy and stability.
Project Insights & Learnings
This project gave me hands-on experience with building end-to-end machine learning pipelines: from collecting raw data to producing a model that could make useful predictions. Working with real-world environmental data presented challenges like noisy inputs, missing values, and changing conditions, all of which pushed me to improve my data handling and model evaluation skills.
Beyond training the models themselves, I focused on making sure the predictions were not just accurate on paper, but stable and interpretable enough to offer practical value for improving wildfire preparedness.