This project explores how guest booking behavior responds to price changes, while identifying distinct demand patterns across customer segments. Rather than optimising price, the analysis prioritises understanding demand elasticity, customer behavior, and acquisition channel dynamics. The core methodology blends price elasticity estimation, demand-driven segmentation, and strategy recommendations, built within a single, reproducible notebook.
NOTE: This project is part of my weekly series in efforts to demystify applied statistical techniques through real-world, project-driven examples, making concepts like causal inference, elasticity modeling, and behavioral segmentation more accessible to practitioners of all backgrounds.
Einstein Ebereonwu • GitHub • LinkedIn
Dataset: Kaggle – Hotel Booking
| Stage | Key Activities |
|---|---|
| Data Quality Review | • Inspected distribution stats to identify implausible values (e.g. zero room price, >8 children). • Detected anomalies using Isolation Forests (outliers ≈ 0.1%). |
| Outlier Handling | • Outliers removed based on iso_score.• All avg_room_price > 400 were also excluded to avoid skew. |
| Missing & Invalid Values | • Replaced avg_room_price = 0 using a KNN regressor trained on clean features.• Detected & removed bookings recorded as 2018-02-29 (non-leap year). |
| Feature Engineering | • Label/Ordinal encoded: meal_plan, room_type, status.• One-hot encoded market_segment.• Created booking_date from year, month, day. |
| Target Variable Design | • Defined demand as number of non-cancelled bookings per day. |
| Exploratory Data Analysis | • Seasonal patterns and demand spikes uncovered. • Room Type 1 and Room Type 4 emerged as most consistently booked.• Clear seasonal peaks in August–October. |
| ANOVA & Correlation | • Identified key demand drivers: room_type, lead_time, avg_room_price, market_segment_Complementary. |
| Elasticity Modeling | • Modeled price elasticity with OLS (linear + polynomial). • Demand for Room 1 is mostly inelastic, turning elastic beyond ~$120. |
| Controlled Regression | • Added controls: lead_time, weekend_nights, month, and market_segment.• Captured a bimodal elasticity distribution hinting multiple behavioral groups. |
| Demand Segmentation | • Segmented Room 1 guests into 3 peak groups using kernel density clustering. • Compared groupwise behavior (lead time, price, cancellations, market segment). |
| Interaction Modeling | • Included interaction terms (e.g. log_price × lead_time, log_price × online) to better explain variance in elasticity across groups. |
Python | pandas | scikit-learn | statsmodels | seaborn/matplotlib | Isolation Forest | OLS Regression | KMeans/KDE clustering
-
Cluster 0- Likely Leisure Travellers- These customers appear to be slightly price sensitive and tend to compare options carefully before booking. They are value seekers who go for the best deals.
- They are organised planners, as they account for the highest frequency of custom requests, and longest lead times (60+ days).
- They also to tend stay the longest at the hotel on average. Eventhough they are not the most frequent guests, they are commitment to their bookings, with low cancellation rates reflect their decision-making certainty.
- They book through both offline and online channels, though like most other guests, the online channel is slightly more favored.
-
Cluster 1- More Flexible Customers- These guests are more flexible in their planning, they do not plan far in advance, compared to Cluster 0.
- They tend to stay longer than leisure travelers, especially during week nights, and their spending is average, neither high or low... However, they are less price sensitive, and are able to pay extra for perceived fairness.
-
Cluster 2- Premium Customers (Likely Business Travellers)- These customers highly value flexibility, they have the highest cancellation rates, shortest booking lead times, and shorter stays, mostly on weekdays.
- This behavior suggests business travel, likely for events or meetings in the hotel’s surrounding area. They are price inelastic, willing to spend significantly for convenience or necessity, and their primary driver seems to be location, flexibility and timing, rather than price or leisure.
-
Cluster 0- Leisure Travellers- Offer targeted promotions such as early bird discounts, tourist attraction arrangements or bundle deals.
- Encourage direct bookings through loyalty incentives and price guarantees... This should encourage them to return as guests, should they decide to visit the city again.
- Highlight customisable options (room upgrades, flexible check-in) to appeal to their planning behavior.
-
Cluster 1- Flexible Guests- Promote mid-range packages and extended-stay discounts.
- Experiment with last-minute deals to encourage shorter planning cycles.
- Market weekday offers to align with their preference for weeknight stays.
-
Cluster 2- Premium/Business Travellers- Provide corporate or executive packages that offer high flexibility.
- Emphasise proximity to business or event venues, fast check-in/out, and premium amenities.
- No generic discounts whatsoever, focus on value-added services like airport pickup, express laundry...
The KDE informed KMeans clustering results for Room Type 4 reveal two distinct traveler segments whose behaviors differ significantly from those in Room Type 1. Unlike the mixed and moderately overlapping patterns seen earlier, these clusters display sharply contrasting characteristics, especially in terms of commitment to bookings, price sensitivity, and planning behavior.
-
Cluster 0- Budget-Conscious, Uncertain Bookers (High Cancellation Risk)- Price-Sensitive: These guests are generally conscious of cost and show tendencies to compare options before committing.
- High Cancellation Risk: They exhibit frequent cancellation behavior, indicating indecision or shifting plans.
- Moderate Planning Behavior: Their booking timeframe suggests some level of planning, but not necessarily strong commitment.
- Diverse Booking Channels: These guests use a mix of online and offline platforms, likely to explore and compare offers.
- Lower Interest in Customisation: Tend to make fewer special requests, indicating a focus on basic needs rather than tailored experiences.
- Lower Demand Volume: They generate fewer bookings overall, reflecting either lower loyalty or less frequent travel.
-
Cluster 1- Committed, High-Spending Guests (Almost Inelastic)- High Willingness to Pay: These guests are ready to spend more for comfort, convenience, or perceived value, showing low price sensitivity.
- Very Low Cancellation Risk: They are reliable and committed, with little to no history of canceling bookings.
- Strong Planning Habits: Bookings are typically made well in advance, suggesting deliberate decision-making and clear intent to follow through.
- High Customisation Preference: These guests frequently request personalised services or specific arrangements, showing high engagement.
- Digitally Engaged: They much prefer online booking channels, reflecting familiarity and comfort with digital tools.
-
Cluster 0- Budget-Conscious, High Cancellation Risk- Require non-refundable or deposit-based bookings to reduce no-shows.
- Limit reliance on this group during peak seasons due to high cancellation volatility.
-
Cluster 1- Committed, High-Value Guests- Promote early bird premium packages and customisable booking experiences.
- Offer loyalty incentives to secure return visits and capitalise on early planning.
- Focus on value-added services rather than discounts (e.g., early check-in, late checkout).
- Working based off my own assumptions that bigger room types are useable as event space. If that holds true, advertising catering & event management services would definately be an ideal move.
While Room Type 1 featured moderately price-sensitive leisure travelers, flexible weekday guests, and premium travellers Room Type 4 stands out due to the extreme polarisation between highly volatile bookers and high-commitment planners.
This analysis highlights how price elasticity modeling and customer segmentation can inform targeted positioning and dynamic pricing strategies beyond mere revenue optimisation. By decoding behavioral demand drivers, hospitality businesses can craft more meaningful, data-driven campaigns and service experiences.
View full analysis on: GitHub | Kaggle Notebook