Filip Piasevoli

Personal blog to post my data science projects as well my digressions on statistics, computer science, machine learning, my career working daily with data (and some pretty brilliant people), and the rock star mathematicians whose work inspires my own endeavors.

This is a personal project where I investigated whether or not one can explain the performance of NFL teams simply by looking at the strength of their draft class. The work was refined following a series of comments and recommendations I received after posting an early version of my work to the NFLstathead subreddit.

This paper takes a closer look at the theory and application of the bootstrap and jackknife methods in statistics. The applications are introduced using test score data by Mardia, Kent, and Bibby (1979) to illustrate the bootstrap estimate of bias, bootstrap estimate of standard error, improved estimates of bias, and jackknife regression amongst other things. This paper was written as a final assignment for Harvard's course 'Introduction to Applied Bayesian Inference and Multilevel Models'.

A link to my Kaggle profile detailing Data Mining competitions I've participated in.

http://xmrrwallet.com/cmx.pwww.kaggle.com/users/93096/filip-piasevoli

Umbrella is a location-based chat application for mobile platforms which we are developing as an entrant into the Boston College Venture Competition (BCVC). We are currently developing for iOS devices and look to expand to other clients. The final round for the BCVC is on April 10th, 2013.

Working with the Massachusetts Bay Transportation Authority, we produced a model of train ridership that can predict how many patrons enter the station on any given day with 95% accuracy. The optimal model for a given station was chosen from an ensemble of models including a simple linear regression, ridge regression, and a random forest model. Our model incorporates historical ridership figures along with information regarding weather and sporting events in Boston. When considering the traffic… Show more

Working with the Massachusetts Bay Transportation Authority, we produced a model of train ridership that can predict how many patrons enter the station on any given day with 95% accuracy. The optimal model for a given station was chosen from an ensemble of models including a simple linear regression, ridge regression, and a random forest model. Our model incorporates historical ridership figures along with information regarding weather and sporting events in Boston. When considering the traffic surrounding Red Sox, Bruins, and Celtics games, we identified the periods of heightened traffic above usual with a metric we call 'lift'. With this knowledge, we advised the MBTA to run trains more frequently at certain times to reduce station congestion throughout the city. Lastly, we identified stations with similar traffic patterns through the use of Principal Component Analysis and k-means clustering. Show less

Using the data we had from our Harvard Capstone Project, we designed an interactive visual tool to explore the question of whether or not Bostonian's commuting habits change when it rains or snows. Being the resilient people that they are, patrons didn't cancel their commutes when it snowed unless there was at least four inches of snow. Rain didn't seem to change how patrons used the T during the week, but any sort of inclement weather on the weekend had drastic effects on ridership patterns… Show more

Using the data we had from our Harvard Capstone Project, we designed an interactive visual tool to explore the question of whether or not Bostonian's commuting habits change when it rains or snows. Being the resilient people that they are, patrons didn't cancel their commutes when it snowed unless there was at least four inches of snow. Rain didn't seem to change how patrons used the T during the week, but any sort of inclement weather on the weekend had drastic effects on ridership patterns. We attribute this change in pattern to the fact that weekend plans are usually more voluntary whereas the weekday commute to work is compulsory. Show less

An investigation into simple analyses a coach may carry out the day prior to a Major League Baseball game in order to gain a better understanding of how an opposing pitcher's strategy may change over the course of the game, batter-to-batter, etc. Most of the project focuses on exploratory analyses as opposed to predictive models. This project was completed with Aaron Zampaglione and Lyla Fadden as a final submission for Harvard's 'Data Science' course.