During the whole period of master at BU, I did lots of personal projects. I summarized these valuable experiences below.

◦ Wrote Support Vector Machine (SVM) and Logistic Regression (LR) algorithms and implemented them on a dataset contains 30 principal components obtained with PCA transformation on AWS(Pyspark) to identify 344 frauds out of ∼ 200k transactions;◦ Used techniques to deal with imbalance dataset, like UnderSampling, OneVsRest SVM and weighted class;◦ Tuned the hyper parameters (regularization term & maxIters) by grid cross validation;◦ Evaluated model by precision, recall and F1 score; the combination of LR and OneVsRest SVM model has the best performance on test set with 89% precision, 80% recall and 84% F1 score;

• Deal with cases from BU students and staffs to help them with their statistical analysis;
 • Solve clients' problems by giving suggestions on experiment design, data collecting, modeling, data analysis and visualization;• Give conclusive suggestions on cases and present to the clients;

◦ Extracted trigger, argument from finance real-time news and classified the news by developing NLP deep learning models: Named Entity Recognition, Joint3EE;◦ Conducted research about competitive products in market to design label schema and discuss application scenarios;◦ Wrote Python scripts to validate the format of the labeled dataset and got widely used for internal employees;◦ Improved model performance by embedding BERT to pre-train model and summarizing wrong classifications to tune hyper parameters (batch size, epoch and LSTM layer, etc.); the best one has 87% F1 score in trigger classification and 93% F1 score in argument classification;◦ Used the result to build Financial Knowledge Graph;

· Finding an appropriate method (Time Series/ Cohort Survival Analysis/ Markov chain Monte Carlo/ Networking Analysis) to project student enrollment situation for each grades and each schools;· Making a budget plan for BPS based on the predicted enrollment situation;

◦ Predicted the days of transformation from Myelodysplastic Syndromes (MDS) to Acute Myeloid Leukemia (AML) by building Support Vector Machine, Random Forest, XGBoost and AutoML models by H2O;◦ Collected the data of 524 patients who match the research purpose out of ∼ 100k MDS patients from Trinity database using SQL;◦ Conducted exploratory data analysis to find out the key factors that lead the transformation occurred;◦ Visualized the Mean Absolute Chap Value Importance, Sharply Value impact and Partial Dependency of all related variables to interpret the XGBoost model with clients;

◦ Conducted exploratory data analysis of ∼ 8k observations on 900 hives with 20 features and generated insights on hives based on beekeeper observations of fertility, productivity, and longevity;◦ Built a LDA model to dig into beekeeper’s visit notes and visualized keywords of them by word clouds;◦ Applied text mining and sentiment analysis on unstructured visit notes to replace some missing values in original dataset and calculate sentiment score(rsentiment) for each beehive;◦ Developed a logistic mixed effects model to predict hive condition based on amount of bees, sentiment score of beekeeper’s visit notes and season;◦ The model informed our recommendations for beekeepers to gather on-site weather and temperature information when visiting beehives; presented the research result with clients;

◦ Predicted video views of a video at the end of its trending by developing a multilevel model on a dataset contains #comment, #like, etc. of YouTube trending videos from Nov 2017 to Jun 2018 in US, UK, CA and IN;◦ Scraped video channel data from website using R(rvest) and added #subscriber as a useful predictor into the model to improve prediction accuracy;◦ Visualized and compared keywords of hashtag and description of videos among 4 countries by word clouds, and interactive displayed the differences by Shiny App (https://yifeng96.shinyapps.io/shinyapp/);

◦ Evaluated the design of website (www.foodeasygo.com), tracked website visiting activities with Google Analytics in a 7-week period and analyzed user behavior in order to promote ‘Lunch’ plan and boost sales;◦ Used A/B testing to compare two versions of advertisement of ‘Lunch’ plan in home page, and determine which version is more effective to increase conversion rate, page views and decrease bounce rate;◦ Provided recommendations of revising the website for manager based on findings;

• Applied R to plot the implicit volatility of historical transaction price of 30+ kinds of commodity in different time scale and found the corresponding distribution to check their stability;• Analyzed the patterns of Candle Chart and calculated the maximum drawdown rate and average drawdown rate to manage the risk;• Applied hedging strategy, found risk-free arbitrage opportunities and kept the stop-loss principle in a short-term simulated future trading with 10 million investment and return of two million in two weeks;

• Acquired data from statistics department and the annual report of Chinese listed companies of 30 provinces;• Grouped the companies according to cities and applied Excel to calculate the concentration ratio and inward foreign direct investment diversity and make regression analysis in R;