Donor / Customer Detector

• To help political fundraisers accurately identify donors and optimize the donation acquisition process, developed a Donor Detector using a dataset of 6.6 million donors.
• Organized and cleaned data, transformed data using label, frequency, and one-hot encoding, performed exploratory data analysis (EDA), generated new features with domain knowledge, and built binary classification models, including Logistic Regression, Random Forest, XGBoost, SVM, and ANN.
• Found Random Forest outperformed other models with the highest precision (.90) and ROC-AUC (.93).
• Provided business insight to optimize fundraising strategies.
• Boosted precision from 56% to 90%, a 60% increase, potentially adding $126 million in annual donations.
• Wrapped the best model as an API using Flask and deployed it on GCP.