Online Shoppers: Who Will Buy?

problem · Only about 15% of visits to an online store end in a purchase. Can a model tell from the browsing session who is going to buy?

approach & results

Trained and compared a decision tree, a random forest, a neural network and XGBoost in Python on 12,330 shopping sessions.
Used recursive feature elimination to keep only the 10 inputs that helped the most.
The tuned XGBoost model was right 91.07% of the time. Since most visitors don't buy, I focused on how many real buyers it caught, which rose from 53% to 58%.
The strongest signal was PageValues, a score for how valuable the pages a visitor looks at are, followed by exit and bounce rates. I recommended offering a live chat or a discount when those warning signs show up.