PROJECTS
ROAD CLOSURES IN DISASTER AREAS
Helped develop a system for scraping social media and news websites to identify road closures and other travel obstacles during natural disasters. This helps disaster victims who stayed in the area to travel safely and assists responders who need to find people in need of help.
I am now working on recreating this work in Python scripts (instead of Jupyter notebooks) to improve its functionality.
WEST NILE VIRUS PREDICTION
Helped predict when and where a mosquito infected with West Nile virus may appear in Chicago using data provided by mosquito traps located around the city. This aids in identifying key locations and times of the year to spray pesticide to prevent an outbreak of the disease.
INSTACART MARKET BASKET ANALYSIS
Using over a gigabyte of grocery transaction data provided by Instacart and the Apriori algorithm, I identified purchasing patterns and associations between items that are commonly purchased together. Due to the size of the dataset, the combinatorial nature of the algorithm, and the limited computing power available to me at the time, I developed a way of trimming the data and folding it continuously to find all associations contained within the trimmed dataset.
I am now working on market basket analyses for a Scottish bakery and an online retailer. I am also making improvements to the function and documentation of the pyfpgrowth Python library.
PREDICTING LEFT-HANDEDNESS
Using survey data, I attempted to predict if a person was left-handed based on their responses to personality questions such as, “I like to take risks” and “I like to go shopping”. I found that answers to these personality questions were not correlated with handedness, and so could not reliably predict left-handedness from this data.
CHRONIC KIDNEY DISEASE PREDICTION
Using biometric data such as blood pressure, sodium levels, presence of anemia, and so forth, I predicted the presence of chronic kidney disease (CKD) in patients with over 98% accuracy. One of the most significant predictors of CKD was the presence of diabetes — a patient with diabetes was 84% more likely to have CKD compared to a non-diabetic patient.
I have found a dataset of diabetic patient biometrics that includes information on the quality of each patient’s sleep throughout the night and plan to start exploring if there is a relationship between diabetes and sleep quality.