Inspiration
We were inspired by the Gaucho Sports Analytics dataset, but wanted to use a dataset more familiar with the group. All of our group members share a mutual interest in the NBA and are passionate about data science and programming. As a result, we decided to meld our twin passions into our project.
What it does
We conducted an Elo ranking analysis of the NBA since 1990. This allows us to evaluate various NBA teams throughout history and make predictions based off their current Elo scores.
How we built it
We used a Python script to scrape the data from BasketBallReference.com for our project. We then ran an Elo ranking analysis on the .csv files from our webscraping in R. We created an ELO system to retroactively assign ratings to every team over the past 35 years as well as use current ELO to predict future results. Used advanced visualizations and time series plots to display results and points of interest.
Challenges we ran into
It was initially quite difficult to scrape the data into the correct format for our project. We also had difficulties creating the appropriate graphs for it as well as difficulty conducting the Elo ranking analysis in R. We had trouble weighing acquiring data manually versus creating/running a webscraper, and the time constraints for both. We were also challenged to find proper parameters that would result in the most accurate model and we also struggled to implement the historical ELO data into modeling future results
Accomplishments that we're proud of
We're proud that we were able to analyze every NBA game since 1990 and get the Elo ranking of teams after each game, consisting of thousands upon thousands of games. We're also proud that we were able to run a simulation of the upcoming NBA games in a reasonable manner. We were able to create a prediction model that nearly matches the current Vegas Sportbooks odds
What we learned
We learned how to web scrape data, conduct Elo analyses in R, how to use time series, and how to run simulations using probabilities. The creation of ELO system was found to be highly representative of real-world team performance. ELO rankings are capable of visually representing the rise and fall of NBA franchises, and ELO can be effectively used to retroactively predict past champions as well as provide predictions.
What's next for Group 44
We shall continue to work to improve the world and follow our interests through data science. As we graduate, we will surely continue to build beautiful and useful things. We can add more parameters in predicting championship outcomes, testing ELO as a metric against in-game statistics and real outcomes, and player analysis using more advanced statistical learning. We can enhance our Elo rating system by incorporating factors such as home-court advantage, player injuries, and other key variables that influence game outcomes.
Log in or sign up for Devpost to join the conversation.