Data Analytics and Science Portfolio by Crystal Lee

About

Hi, I am a savvy technical analyst with 4-year experience in data analytics and 2-year experience in fast-paced startup environments. Proficient in SQL, Python, dbt, and various BI tools, such as Tableau and Metabase.

This portfolio is a compilation of all the data science and data analysis projects I have done for work, academic, self-learning and hobby purposes.

Skills

Database: BigQuery, MySQL, PostgreSQL, MongoDB
Programming: SQL/dbt, Python, R, JavaScript
Data Engineering: dbt, Airbyte, Fivetran, Airflow, Spark
BI Tools: Tableau, Metabase, Google Data Studio
Product & Marketing: Segment, fullstory, Jira, Google Ads, Google Analytics

Data Apps

1. Chatbot App

This Chatbot App built on Acho uses conversational AI to create human-like interactions with users by leveraging OpenAI’s text completion API and Python model.

2. Sales Simulation App with Python

This App powered by Python utilizes the Monte Carlo model to forecast sales for a business. Monte Carlo methods are a class of computational algorithms that rely on repeated random sampling to obtain numerical results.

3. Digital Advertising Performance Tracking App

The digital advertising performance tracking app is a tool that helps marketers and advertisers track and measure the effectiveness of their digital advertising campaigns.

4. Data Request System

A data request system functions enable users to submit requests for data, dashboards, or ad-hoc reports. Data owners have visibility of the requests and can address them accordingly.

5. Self-service Pivot Table App The pivot table app allows users to create and analyze pivot tables, which are a type of data summarization tool used in spreadsheet programs like Microsoft Excel and Google Sheets.

6. Virtual Movie Recommendation Assistant

Purpose: A chatbot to help users find the movie of interest, and provide recommendations based on the chosen movie
Methods: Web Scraping, Finite State Machine, Recommender System, Web Application
Tools: Python, Plotly Dash

7. Loan Analytics Calculator and Dashboard:

Purpose: A web application to calculate monthly payment for multiple loans in tables and charts
Methods: Data Visualization, Web Application
Tools: Python, Plotly Dash

8. Route Planner and Network Analysis

Purpose: A web application to optimize travel routes and offer information about cities along the route
Methods: Network Analysis, Data Visualization, Web Application
Results: Python, Plotly Dash

[Top]

Python and R

1. Clothing E-Commerce Reviews Sentiment Analysis

Purpose: Understand customers' attitudes toward the business
Methods: Naïve Bayes Algorithm, Machine Learning (Classification)
Results: F1 score achieved 78.1% and identified 53 keywords for classifying positive and negative reviews

2. Exploratory the Business Operation of an E-Commerce

Purpose: Discover insights from the status quo
Method: Descriptive Statistics, Cohort Analysis, Visualization
Result: Strategies works very well after August and increasing the retention rate is indeed helpful to raise revenue

3. E-Commerce Customer Segmentation

Purpose: Identity current customer groups
Methods: RFM analysis, K-means, Machine Learning (Clustering)
Results: Segmented consumers into 5 groups and found a potential risk that sales heavily relies on a few customers

4. Channel Attribution Modeling in Digital Marketing

Purpose: Recognize channels which contributes the most sales
Methods: Markov Chain, Visualization

Results: In the 5 channels, Facebook and Paid Search contributes 54.4% conversions, whereas Instagram has the highest conversion rate

5. Predictive Modeling for Bank Telemarketing

Purpose: Find out the best times to call the right customers to promote a term deposit
Methods: Classification, Logistic Regression, KNN, Random Forest
Results: Implemented several machine learning models and selected the best performing random forest model which had the best precision score

6. Market Shocks of COVID-19

Purpose: Explore how industries react to the market crash due to COVID-19
Methods: Web Scraping, K-Median Clustering, Visualization
Results: The energy sector suffers from a considerable decrease in the stock price, but technology, consumer products, and healthcare are relatively robust

7. Research on COVID-19 Comorbidity

Purpose: Study which diseases probably co-occur within COVID-19 patients
Methods: Association Analysis, Visualization
Results: Summarized top 20 rules with the highest lift and further explore if causality exists between diseases

[Top]

Natural Language Processing

1. Naïve Bayes Algorithm

Developed and implemented a multinomial Naïve Bayes classifier using bag-of-words features from scratch

2. N-gram Language Model

Created unigram and bigram language models to solve the jumbled sentence task, that is, to find which sentence is a real sentence out of 10 jumbled sentences

3. Part-of-speech Tagging with Hidden Markov Models