This was a consulting project during the Insight Data Science fellowship.
The objective here was to present a probabilistic approach for ad bidding by determining which users were most likely to install the app if they were shown an ad.
There are 4 notebooks that denote these steps:
- Data Extraction
- Feature Exploration
- Predictions
- Probability Model
The data is propietary and can't be shared.
A main challenge to this project was that the classes were imbalanced, 99.5% of the people in the data did not install the app when they saw an ad. To combat this several approaches were used and the one I stuck with is an approach that takes into account the prior conversion rate into account.