Keeping Customers Content
Dataset: Santander Customer Satisfaction dataset
Goal: Predict dissatisfied customers without detailed features and background.
Dataset: Anonymized dataset with 370 features for 76,020 customers
Solution: For the golden ensemble, Firefly Lab selected four Random Forest and one Gradient Boosting models. During preprocessing, Firefly Lab drew on Decision Tree and Logistic Regression for feature stacking, then engaged in feature selection for the Gradient Boosting model only, using the ExtraTrees Classifier.
Firefly Lab Results: 83.418%
Rank: Exceeded 1st place of 5123 competing teams
(Firefly’s score outdistanced the winning entry by the amount that the winner exceeded the 2734th place entry.)
Data Scientist Time: 20 minutes