README.md

Satellite_Images_Embeddings

Embeddings:

Embeddings of 1024 features generated using the satellite images of the top 5 municipalities with most dengue cases in Colombia. The embeddings were generated usign dimensionality reduction methods such as:

1. Supervised Contrastive Learning:

• Code to create the model: Contrastive_learning.ipynb
• Code to generate the embeddings: Features_Contrastive_learning.ipynb

2. Variational Autoencoder

• Code to create the model: Variational_Autoencoder.ipynb
• Code to generate the embeddings: Features_Variational_Autoencoder.ipynb

3. Autoencoder

• Code to create the model: Autoencoder.ipynb
• Code to generate the embeddings: Features_Autoencoder.ipynb

The embeddigns can be found in the folder Embeddings/ with the names:

• Features of Supervised Contrastive Learning: embeddings_contrastive_learning_1024features.csv

• Features of Variational Autoencoder: embeddings_vae_1024features.csv

• Features of Autoencoder: embeddings_autoencoder_1024features.csv

All labels used to train the models can be found in the folder: Dengue_dataset/

• Labels with dengue cases per inhabitans for regression: cases_per_inhabitants.csv
• Labels with peak or not peak of dengue for binary classification: binary_classes.csv
• Labels with increase, decrease or stable dengue cases across weeks for multi-class classification: multiclass_labels.csv
• Labels with dengue cases across weeks and other sociodemographic/socioeconomic variables per municipality: dengue_data_all_municipalities.csv

Machine Learning models with the dataset:

1. LSTM for regression of dengue cases using the embeddigns.

2. LSTM for multiclass classification using the emebeddings.

3. Binary classification + Contrastive learning.