Xinyan Ou Email & Phone Number

Customer Churn DetectorTo avoid revenue loss from customer churning in a telecom company, built a Customer Churn Detector based on binary classification models, and developed an Insight Engine for Marketing Team to generate sale strategies. Collected and preprocessed data, conducted EDA and feature engineering, used SMOTE to handle the imbalanced dataset, built multiple classification models including Logistic Regression, Decision Tree, Random Forest, SVC, XGBoost, Gradient Boost.Achieved a Recall score of 0.88 with Logistic Regression model, 3 times more effective than before to locate churning customers. Achieved a roc_auc score of 0.78 with Random Forest model, used SHAP to identify key factors, provided insights for marketing strategies.

Greater New York City Area

Real-time Decision Maker of Material-Handling RobotsTo lower the production loss due to machines' random down events during manufacturing process, built a Data-driven Real-time Decision Maker to control the operations of the material-handling robots.Built a manufacturing simulation environment, applied multiple Machine Learning methods, including Reinforcement Learning, Approximate Dynamic Programming and Online Task Allocation, to find the best moving strategies for the material-handling robots.ML strategies effectively reduced the impacts of machines’ random down events, reduced 11.18% production loss in average, comparing with traditional strategies.

Michigan

Root Cause Identifier for Product QualityTo lower the product failure rate of a crankshaft production line, built a Root Cause Identifier based on binary classifier models.Collected and cleaned raw data from factory, filtered features from over 600 features use PCA and conducted feature engineering based on domain knowledge, built a series of machine learning models including Logistic Regression, Decision Tree, Random Forest, SVC, XGBoost, Gradient Boost.Achieved a ROC-AUC score of 71% with Logistic Regression Classifier, found the most important factors that impact product quality. Detected two important optimal temperature range, in which the failure rate could be lowered from the current 2.97% to 2.07%.

Doctor Of Philosophy - Phd, Mechanical Engineering

Master'S Degree, Mechanical Engineering

Bachelor'S Degree, Mechanical Engineering