Airbnb Price & Review Score Analysis (Boston & Seattle)

This project analyzes 2021 Inside Airbnb listing data for Boston and Seattle to answer two sets of questions: what drives the nightly price of a listing (neighbourhood, property type, capacity, amenities), and what drives its review score (host responsiveness, acceptance rate, superhost status, profile completeness). The work follows the CRISP-DM process end to end, from business understanding through data preparation, modeling, and evaluation.

Data cleaning covers the usual mess of real-world listing data: percentage and currency strings coerced to numeric, boolean-like t/f columns encoded, NaN imputation, and outlier treatment via IQR-based thresholds (which cut the price column's skew from ~21 to ~1.6). A custom parser turns the free-text amenities column into one-hot features for the most common and a few hand-picked amenities (pool, gym, parking, etc.), which then feed a correlation-based feature selector.

The price side concludes in a linear regression model, with its performance compared across different correlation thresholds for feature selection, across price bands, and across the two cities separately versus combined — the model doesn't improve by splitting per city, suggesting the underlying price dynamics are similar in both markets. The review-score side stays exploratory: none of the host-behavior signals tested showed a strong relationship with review scores.

Full findings are written up in the linked Medium article; the notebook itself is structured with markdown commentary at each CRISP-DM step so it can be read as a standalone report.