Skip to content

Projects

Airbnb Superhost Classifier

Break Through Tech

Logistic regression predicting superhost status, tuned with GridSearchCV to 0.82 ROC AUC on held-out data.

Trained a logistic regression model in scikit-learn to predict whether an Airbnb host holds superhost status from listing and review features. Ran GridSearchCV with 5-fold cross validation across 10 regularization values to select the hyperparameter, then compared the tuned model against the default using confusion matrices, precision-recall curves, and ROC curves. Reached an ROC AUC of 0.82 on the held-out test set, with tuning producing almost no gain over the default, and used SelectKBest to confirm review volume and host response rate carried most of the signal. Serialized the final model with pickle and reloaded it to verify it predicted correctly after deserialization.

Enter SelectEsc Back