Audit and clean
Profile every column, remove duplicate or invalid records, standardise formats and units, investigate missingness and outliers, and record why observations are corrected, retained or excluded.
← Selected work
Team leader · Current project · Work in progress
Airbnb market intelligence · Listing analytics
Airbnb Market Intelligence
I'm leading the team's analysis of how listing characteristics influence price and booking popularity. The objective is to clean the data comprehensively, investigate every usable variable and let the evidence—not our first intuition—determine what enters the model.
Analytical process
Clean broadly. Test systematically. Model what the evidence supports.
Rather than pre-selecting only the variables that seem obvious, I am structuring a reproducible process that audits all non-constant fields with usable variation, examines their relationships with price and popularity, and documents every modelling decision.
Screen every usable field
Test all variables with meaningful variation against price and popularity, including obvious fundamentals and less intuitive fields such as minimum-night requirements or availability rules.
Investigate interactions
Examine non-linear and conditional effects—especially season × location—to avoid assuming that one driver has the same influence across every market and period.
Build comparable cohorts
Group listings with genuinely similar characteristics, estimate price and popularity from relevant peers, then validate whether the model remains stable outside the development sample.
Driver investigation
Separate intuitive relationships from relationships the data can defend.
How much of price and popularity is explained by location, season and property type?
Does a higher minimum-night stay reduce demand, or signal a different listing and guest segment?
Does seasonality matter differently across locations rather than producing one market-wide effect?
Which nearby listings are similar enough to provide a credible pricing and demand benchmark?
A practical example
Location and season may be inseparable.
IntuitionA listing near a seasonal attraction—such as a lake that is visually distinctive only in summer—may command high prices and strong demand during a narrow peak.
Interaction testThe same location may be substantially cheaper and attract fewer guests outside that season, so separate location and season variables may miss the real relationship.
Model responseTest the season × location effect and compare the listing with peers that share the same property, market and timing characteristics.
Business applications
Turn listing data into explainable pricing and demand decisions.
Estimate a defendable range
Use genuinely comparable listings and the most relevant drivers rather than one market-wide average.
Anticipate changing popularity
Identify when season, location and booking constraints combine to strengthen or weaken expected demand.
Explain performance differences
Separate structural listing advantages from temporary market effects before comparing properties or recommending action.
Team-led project · In progress
Build the evidence before choosing the model.
- Complete the column-level data-quality audit and cleaning log.
- Define defensible measures of listing popularity alongside price.
- Test non-linear relationships and season × location interactions.
- Select comparable cohorts and validate the model out of sample.
- Coordinate team responsibilities, review evidence and keep the final narrative consistent.