John Leraas

Project Portfolio

Haiti Disaster Relief

This project was done as part of a class on machine learning and used actual satellite data on Haiti following the earthquake in 2010. Coded in R, utilizing the following classification models: logistic regression, linear discriminant analysis, quadratic discriminant analysis, K-nearest neighbors, penalized logistic regression, random forest, and support vector machines.

AB Testing: Frequentist vs. Bayes

A/B Testing is an extremely common practice in web optimization, clinical trials, and a range of other applications. While the Frequentist Approach is the most commonly employed method in A/B testing, a Bayesian Approach offers many advantages including: requiring fewer observations, a more straightforward analysis, and more intuitive results. Additionally, a Bayesian approach allows multi-armed bandit strategies to be employed during the experiment to reduce opportunity cost.

Subreddit Exploratory Text Analytics

Implementation of various NLP techniques to compare and evaluate a select subset of 20 subreddits. Organized text into Ordered Hierarchy of Content Object Structure and applied principal component analysis, topic modeling, sentiment analysis (NRC Word-Emotion Lexicon and Valence Aware Dictionary for sEntiment Reasoning), word embeddings, and similarity measures (cosine similarity, Jensen Shannon Divergence) with hierarchical clustering.