← Portfolio
▲
Neural Network Wildfire Risk Classification
Eric Solak
· MLP trained on NASA FIRMS + Open-Meteo data · PyTorch
118,858
training samples
94.3%
test accuracy
ⓘ Model details
Code ↗
×
About this model
Multi-layer perceptron · PyTorch
94.3%
Test accuracy
0.939
Macro F1-score
118,858
Observations
Architecture & training
4-layer feed-forward network with batch normalization and dropout regularization
Weighted cross-entropy loss to handle class imbalance across risk categories
L2 weight decay and early stopping against overfitting
Data & features
NASA FIRMS fire detections joined with Open-Meteo weather history
17 engineered environmental features, including temperature–humidity and wind–FWI interactions
Outputs one of three risk classes with a confidence score — this map shows a 2,001-point sample of predictions
GitHub ↗
Technical report ↗
Live API ↗
Loading model predictions…