Parcel Price Prediction
![](/project/parcel-price/featured_hu4eb33b2b5dbcabca3e9d86c3de56f563_372814_720x2500_fit_q75_h2_lanczos_3.webp)
Using Boston’s property assessment data, I built a prediction system for the price of a land parcel using its physical attributes. Stakeholders who might use the predictions are the city of Boston itself (to determine property tax) and homeowners (to find a fair value for their purchase). I used pandas to manipulate csvs, sklearn to preprocess (removing outliers, imputing missing features) and cross-validate, and finall LightGBM to fit learner. On single-household residences, the system is reasonably accurate: the average absolute percentage error is about 25%. The most important predictive feature is the total living area.