// the find
seehiong/hdb-price-predictor
A Streamlit web app that predicts Singapore HDB resale flat prices using a pre-trained XGBoost model. Includes an interactive transaction map, light/dark theme toggle, and Docker support for easy deployment.
A Streamlit app that predicts Singapore HDB resale flat prices using a pre-trained XGBoost model, with an interactive Folium map of recent transactions. Aimed at Singapore property researchers or anyone learning how to ship a simple ML prediction app with a web UI.
The Docker setup is clean and gets you running in two commands. Shipping the model binary and scaler directly in the repo removes the 'where do I get the model?' friction that kills most demo ML projects. The Folium transaction map with clickable markers per block is genuinely useful for context — seeing what nearby flats actually sold for matters more than a point prediction alone. Data source is cited (data.gov.sg) and dated, so you know exactly how stale the training data is.
The pre-trained model is a black box — there's no training notebook or feature importance output in the repo, so you can't tell which features drive predictions or retrain when new data drops. The interactive map is iframed from an externally hosted HTML file on GitHub Pages; if that URL breaks, the map just disappears with no fallback. No model versioning or metadata (when was it trained, on what date range, what's the RMSE on a holdout set) — you're trusting the binary blindly. The related blog post link goes to a KServe/OKE deployment post that has nothing to do with this repo, which suggests the README is copy-pasted from another project.