Airbnb Pricing

What drives nightly price? EDA, ANOVA, SHAP and hierarchical market segments across 3,982 listings.

Year
2025
Course
ISE 535 Data Mining · USC
Listings
3,982
Stack
SHAP · Gower distance · Hierarchical clustering · statsmodels

A data-mining study of 3,982 Airbnb listings from one metropolitan area, with two goals: understand what drives nightly price, and find natural market segments (budget, family, luxury).

Workflow

  1. Explore: distributions, outliers, and rates by property type, capacity and season.
  2. Test: ANOVA and t-tests on capacity, seasonality and cleaning fees.
  3. Model: linear regression vs. decision trees, compared on R² and MAE.
  4. Explain: SHAP and partial dependence plots for global and local effects.
  5. Segment: Gower distance with agglomerative (hierarchical) clustering, projected with PCA, and each segment profiled.
A singer at a microphone, a slanting orange stage light across the face
Airbnb PricingCover: N°01, Procreate
A guitarist in an LA cap looking down at the guitar neck, under blue light
Airbnb PricingCover: N°02, Procreate
A single red rose in slanted window light, a blurred leaf shadow on the wall
Airbnb PricingCover: N°03, Procreate
Three sunflowers on a black ground covered in loose grey scribbles
Airbnb PricingCover: N°04, Procreate
A charcoal drawing of a hand holding a rose and baby’s breath
Airbnb PricingCover: N°05, Charcoal
A black-and-white calla lily wrapped in loose looping lines, with pixelated blocks
Airbnb PricingCover: N°06, Procreate
A black-and-white sketch of a curly-haired guitarist leaning into a solo
Airbnb PricingCover: N°07, Procreate
A dark grey rose emerging from a brushed charcoal background
Airbnb PricingCover: N°08, Procreate
A candle melting in an open hand, lit against the dark
Airbnb PricingCover: N°09, Procreate
A sun sinking into dark water, painted with glitchy horizontal brush streaks
Airbnb PricingCover: N°10, Procreate

About

I’m an engineer with a background in data analytics and systems engineering. I like building things close to the data: recommendation systems, local-first assistants, analysis pipelines, and interfaces that make a model’s output easier to read.

When I’m not coding, I paint, mostly in Procreate and sometimes in charcoal. This site keeps both in one place.