Predicting New User Bookings Anaelia Ovalle, Michael Liston, Brent Rucker
Predicting New User Bookings
Anaelia Ovalle, Michael Liston, Brent Rucker
Table of Contents
I. Introduction to Project & Goal
II. Data Pre-Processing
III. Models
IV. Results
V. Discussion
Data Sources
Libraries
GoalUsing a dataset of 15 basic features, predict
where the user will make their first booking
Country Destination
12 countries
14 predictors possible
Data Pre-Processing
1. Observe all distributions
2. Identify NA’s and handle NA
3. Varied Training and Testing
4. Date Feature Extraction
5. One-hot encode categoricals
a. 10/14 predictors categorical
6. Binning
Age Feature Imputed by Mean
Modeling with Multi-Class Classification
● 16 Models○ Decision Trees○ Random Forests○ AdaBoost○ QDA○ KNN○ XGBoost○ SVM○ Neural Network
How Many Trees?
Sample Code with Tuning Parameters
Feature Importances
Results
Discussion
Best Accuracy: Random Forest
Best Accuracy != Best Model
Best Precision: Gradient Boosting
Challenges
Access to more structured data
More sophisticated imputation methods
Evaluate more models
Motives of Airbnb
Time
Business Applications
Precision vs Recall
Use Recall
Increase FN
Increase Spam
Negative impact on Reputation
Use Precision
Decrease Spam
More bang for buck
Smarter Decisions
Thank you