Projects

Activision Data Science

Role — Data Scientist, People Analytics

Apply predictive analytics and state of the art Natural Language Processing techniques to improve both employee (micro) and company-wide (macro) infrastructure.

Commonly used skills:

• LSTM
• Word vectorization
• Text Processing
• Multivariate regression/forecasting techniques such as ARIMA

Lane Detection

Challenge

Apply computer vision and image processing techniques to successfully detect lane lines from a forward facing dash cam.

Techniques:

• Open CV
• Hough transforms for crude line detection
• Image segmentation using masking and polygon stenciling
• Thresholding

F-MNIST Image Recognition

Description

The F-MNIST dataset consists of 60,000 black and white images of clothing, all of 28x28 resolution.

Challenge

Use multi-class classification techniques to accurately classify images of clothing.

Techniques:

• Stochastic gradient descent
• K-Nearest-Neighbors analysis
• Principal component analysis
• Error analysis using ROC/AUC curves and confusion matrices
• K-Fold cross validation methods

Regression Analysis for California Housing Market

Description

24,000 row, 13 column dataset consisting of various parameters including median house price, median size, number of bedrooms, median income, etc.
Each sample is a district (block) consisting of anywhere between 20-400 homes. Dataset parameters consist of averages of each district.

Challenge

Clean and analyze all data. Find emerging trends in housing and why these trends may or may not persist.
Techniques included:

• Correlation matrix
• K-Nearest-Neighbors Analysis
• Principal component analysis
• Automating data-cleanup using custom transformation pipelines
• K-Fold cross validation methods

NBA Statistic Analysis

Challenge

Using web-scraping, pull data from an NBA statistics site and create a dashboard to view data.

I applied various machine learning and analysis techniques which included:

• Python Dash library
• HTML/CSS styling
• Matchup analysis using probability modeling