Projects

Multiple Projects for Data Analysis in various industries(Retail, e-Commerce, Real Estate, Manufacturing, Transportation, etc)

Twitch API Analysis(Python):

• Understood project manager's requirement for finding important developer/API endpoints
• Cleaned and extracted features of daily_logs from Nov 2017 to Feb 2018, , mapped with application_metadata
• Analyzed and visualized daily_log for trends, correlationships between features

Zillow House Price Prediction(R/Python):

• Exploratory data analysis; Data Mining; Imputation of Missing Data; Feature Selection& Generation;
• Built machine learning Models(Linear Regression, Random Foest, XGBoost) to predict next season Zillow Price; Tuning Parameters and Modifying models.
• R Shiny for visualization of transaction and geometric data(How properties and their price vary from city to city in CA?).

Instacart Online Grocery Store Customer Reorder Prediction(MySQL, R, Python)

• Built relational-database and ERD to clarify relationships between customers, retailers and products, normalized raw data( MySQL)
• Including EDA, Customer Segmentation(demographic, historical purchase behavior analysis, product-based segments)
• Query in MySQL, visualization in Tableau to provide insights and recommendations to Instacart team

Implementation of Mixed base-learner Adaboost, modified by Genetic Algorithm(R)

Self-written mixed weak learner of Adaboost with feature selection using Genetic Algorithm on Real-world Binary Classification problems:

• Select 4 base weak learners among 12 learners by grid search, trials and evaluations
• Implemented AdaBoost algorithm with updating weights of both training dataset and learners(respectively) in each iteration
• Applied GA in weak learner combination selection, tuning parameters such as crossover rate, mutation rate, elicit status, etc. to optimize the final model performance.
• Reduced overall model complexity by 75%, without decreasing model preformance while increase model's interpretability
• Conduct parameter tuning and feature engineering, increased 6% of prediction accuracy

Uber Rider Behavior Analysis(Python)

• Data cleaning, extraction, EDA of NYC uber rider/driver behavior
• Time series analysis, feature engineering
• Setting 'Churn label' based on different requirements
• Built and modified rider churn prediction models (Logistic Regression, Random Forest) using Sklearn
• Preformed Cost Benefit Analysis of methods in new user acquisition and potential churning user retention

Real Estate Estimation of best investment area in NYC(Python)

• Conducted data mining, feature extraction on Zillow historical house price estimation and Airbnb short term rent price, based on ad-hoc business target: Find the best investment area for short-term leasing
• Preformed data munging on particular features with multiple units in different datasets, wrote functions to link data together in a scalable way to allow new data append
• Created specific metadata and metrics, such as Cap Rate/Occupancy Rate to refine and better understand business goal
• Successfully target best investment area in NYC based on defined metrics and trend prediction

Water Usage Capcity Analysis and Prediction(SAS)

• Including feature generation/selection, transformation(boxcox, );
• feature selection using various techniques/criteria: C_p, stepwise) in building Linear Regression Model
• Checking assumptions, giving diagnostics using metrics such as studentized residuals, Cook's D, hat matrix diagonals, toleance, VIF, etc.
• Making predictions based on the selected model.