AirROI Investment Analytics

Project Details

Overview & Team Collaboration

Developed for educational purposes through NC State's Institute for Advanced Analytics, this project was built using public data from AirROI, a platform that aggregates short-term rental metrics to market analytics tools for property investors. Operating as independent external consultants, my team was tasked with conducting an open-ended analysis of AirROI's regional datasets to build a sellable investment guide and dynamic pricing tools for aspiring Airbnb hosts. We have no formal relationship with or endorsement from AirROI.

We structured our analysis around three guiding questions:

1. Where & Why? – Identifying geographical hotspots with the highest revenue potential relative to hotels.
2. What to Optimize? – Quantifying the nightly rate premiums of specific amenities and booking attributes.
3. How to Price & Market? – Designing a machine learning model to dynamically predict prices and assessing the visual quality of listing photography.

---

Methodology & Machine Learning

Our team developed three core analytics pipelines:

• Hotel vs. Airbnb Benchmarking: Compared comparable Airbnb listings (≤ 4 guests, ≤ 2 beds, 1 bath) against local Federal Per Diem (FPD) Lodging Rates (Hilton-equivalents) to classify markets as Hotel-Competitive (e.g. Raleigh, Durham) or Desirable Hotspots (e.g. Gatlinburg, Carolina Beach).
• Spatial Neighborhood Modeling (KNN): Designed a K=5 K-Nearest Neighbors spatial model using spherical haversine distance to calculate spatial lags and map localized revenue hotspot zones.
• XGBoost Dynamic Pricing Model: Engineered a feature-driven model with 300+ columns (cancellation policies, booking options, seasonal variables) and text/image vector embeddings. We selected an XGBoost Regressor with a Log-Link Gamma objective to handle the skewed distribution of nightly rates.

Our champion model achieved a Mean Absolute Error (MAE) of $36.58 and an R² score of 0.8456, capturing 85% of the pricing variance across the sub-markets.

---

Key Findings & Actionable Insights

1. Booking Attributes & Price Premiums

We isolated how operational listing attributes directly impact average daily rates (ADR):

• Strict Cancellation Policies: Stricter policies correlated with higher nightly rates. Super Strict 60-day policies command an average of $414/night compared to Firm ($235) and Flexible ($182) policies.
• Instant Booking: Removing booking friction yielded a significant $88/night premium ($306/night for Instant Book vs. $218/night for Request to Book).
• Pets Allowed: Allowing pets drove a +$26/night premium ($255 vs $229).
• Superhost Status: Regular hosts averaged higher nightly rates ($274/night vs. $214/night for Superhosts), indicating that Superhosts prioritize high occupancy and review accumulation over top-dollar nightly pricing.

2. Photography & Cover Photo Residual Analysis

By analyzing our model's pricing residuals (Residual = Actual - Predicted), we isolated the visual factors that cause listings to overperform or underperform expectations:

• Overperforming (High Residuals): Characterized by a strong "color pop," high-quality lighting, and cover photos that show a holistic view of the property or clearly highlight premium amenities (like private pools and hot tubs).
• Underperforming (Low Residuals): Associated with dim lighting, generic stock resort images, and poor framing (e.g., cut-off furniture or empty forest views).

---

Interactive Dashboards

To put these tools in the hands of investors, we created two interactive Streamlit applications. You can access the live dashboards using the links below:

• Market Spatial Analysis Dashboard: Visualizes K=5 nearest-neighbor pricing spheres and lets investors zoom into specific sub-market coordinates to locate high-performing micro-locations.
• Amenities & Attributes Impact Dashboard: Lets hosts filter by bedrooms and bathrooms to isolate the exact dollar rate difference and saturation rate of specific amenities.