Portfolio
Classification and Hypothesis Testing:
Predicting Hotel Booking Cancellations
​
Objective:
Hotel cancellations and no-shows are an ongoing revenue challenge as guests today can easily cancel bookings, even at the last minute, thanks to flexible online booking options, presenting hotels with lost revenue from rising cancellation rates if unmanaged. To address this growing cancellation problem for INN Hotels Group, a machine learning solution applied to their Portugal hotel booking data could help predict and manage cancellations by identifying key drivers, developing a predictive model to forecast cancellation risk, and recommending data-driven cancellation and refund policies that protect revenue.
​
Key Findings: Features in order of importance
-
Lead time: The older the booking, the higher the probability of cancellation
-
Average price per room: Bookings during high room rate time frames are more likely to be canceled.
-
Number of special requests: Bookings with no special requests are likelier to cancel.
​
Model Results:
​
Recommendation Systems:
Recommending Movies to Users
​
Objective:
Streaming services like Netflix offer vast libraries of films to viewers. Developing recommendation algorithms that suggest movies tailored to each user's tastes and history could enhance customer satisfaction. More satisfied users may lead to increased revenue for these platforms.
​
Models:
-
Rank-based Recommendation System
-
User-based Collaborative Filtering Recommendation System
-
Item-based Collaborative Filtering Recommendation System
-
Model-based Collaborative Matrix Factorization using SVD
​
Results:
The Model-based Collaborative Matrix Factorization using SVD model provided the best overall results measured by precision and recall.
​
Precision: 0.738
Recall: 0.525
Visualizations
