This is a repository for my personal portfolio page.
All project code are hosted on Kaggle.
Paper on findings in the end of the notebook.
Exploratory data analysis for regional sales correlation and categorical variables/interaction significance. Feature reduction on baseline model, model diagnosis and improvements. Two-part model to accomdate zero-inflated, power-law-like distribution of North America sales and model performance evaluation.
Classify critical mineral presence using xgboost and autoencoder embeddings created from global geochemical soil, whole rock and terrain data. (No kaggle codebook for ethical reasons)
Most widow earns less than $50k/year.
Grid search experiments for runtime v.s. accuracy trade-off.
Recommending crops for African countries.
Satellite embeddings as input features.
Explore climate embeddings: finding future locations (purple) with similar climate to current location (green blue).