Spring 2018 Projects

Arable

Team Members: John Gallagher, Yihe Yang

Farmers fear heat waves and must take precautions to avoid them. Arable produces an IoT device called Pulsepod that can track local weather parameters such as ground temperature, humidity, etc. Using Arable data, students built predictive models that can forecast the next heat wave. Team members used time series, LSTM (Long Short Term Memory), ARIMA, KNN, Random Forest, and SVM to make predictions.

Brand Network

Team Members: Matt Trudeau, Karan Vombatkere, Josh Kolodny

For social media advertising, marketers must determine optimal bid types to maximize the impact of their campaigns. The goal of this project is to create models and methods that can suggest ideal bid types in real time. These models help optimize KPIs during campaign setup, based on marketing goals and criteria. Team members applied z-score normalization and One-Hot Encoding to preprocess data, then employed Neural Network and Logistic Regression to predict bid type based on historical advertising data and campaign attributes.  

IEEE

Team member: Yue Zhao

Technical documents published by the Institute of Electrical and Electronics Engineers (IEEE) are typically peer-reviewed. Traditional peer review processes can be laborious and challenging, especially for large submissions. This project aims to create a machine learning-based, automated reviewer recommendation system for manuscripts. To achieve this goal, the student performed exploratory analysis, computing a word cloud that could find the most frequently-used words in abstract, using LDA to determine topic composition.

Origent

Team Members: Hehua Chi, Brooke Hamilton

The Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS) is a clinical score for evaluating the functional status of patients with Amyotrophic Lateral Sclerosis (ALS). Using ALS patients’ medical information (with records of use of the drug Riluzole) students tracked changes in the ALSFRS scores for patients taking Riluzole, comparing their scores to those who didn’t take the drug. Team members employed cluster analysis (PCA, K-means, and WCSS) to explore the data set. They then applied regression, random forest, and gradient boosting to the data to predict the rates of change in patients’ ALSFRS scores.