Chen Lin Email & Phone Number

Genome-Wide Prediction of IMF2 Data 1. Developed the SREML (stochastic restricted maximum likelihood) method by combining REML method and Bayes Rule and implemented it for integrated study of trait phenotypes and genetic markers in R.2. Achieved the improvement of predictability compared with best linear unbiased prediction (BLUP), ridge regression, LASSO and Bayesian approaches.3. Improved computational efficiency by applying HAT method that provides the cross-validation results using the whole sample only once.

Breast Cancer Detection 1. Applied logistic regression, KNN (k-nearest Neighbors), SVM (support vector machine), neural networks and decision trees on detection of breast cancer in R.2. Constructed exploration data analysis to select features that discriminate two groups significantly.3. Evaluated the model performance by cross-validation method and provided ROC curve for logistic regression.4. Reduced the misclassification rates or false positive rates (FPR) by implementing ensemble learning which uses multiple learning algorithms to obtain better predictive performance.

Handwriting Recognition 1. Applied SVM with linear kernel to build a handwriting recognizer to solve multi-class classification problem in MATLAB.2. Reduced the computational demanding by using PCA (principle component analysis) to decrease the data dimension.3. Improved the model performance by replacing linear kernel with RBF (radial basis function) kernel.

Worked in Seller Risk Modeling team at Paypal which maintains statistical models that are used to drive risk actions for seller risk management including hole, reserve, caps and teammate reviews.Worked on full cycle of model development process including data preparation (using Teradata SQL and SAS), variable selection, scoring model development, and model validation and diagnostics.Used SAS/MACRO and Teradata SQL to isolate non-repeating risk transactions to save teammates time.

Phd Candidate, Statistics

Master'S Degree, Statistics