Inspiration 🌟
Our journey began with a simple question: How can we accurately rank eSports teams? Drawing parallels with traditional sports, especially football and FIFA's methods of team ranking, we embarked on a mission to create an effective ranking system for eSports teams.
Starting Simple 🛠️
Initially, we wanted to lay the groundwork using the minimum requirements. We adopted a basic formula, inspired by the documentation provided about how team rankings work in football. Leveraging the win rate of each team and the ELO they played in, we were able to segment the teams into categories. However, while this method could determine superior and inferior teams, the internal rankings within each segment didn't feel accurate.
Diving Deep into Data 📈
With the goal of refining our rankings, we ventured into analyzing extensive game data. This data was vast, with individual game files often stretching to 100-150 MB. This posed our first major challenge: the need for data cleaning.
Simultaneously, to enhance our formulaic ranking system, we dabbled with the ELO system. Ranking different leagues and assigning them scores gave us surprisingly accurate results for top-tier teams.
Cleaning and Refining 🔍
As we dug deeper into the game data, we adopted an aggressive data cleaning approach. By focusing on end-game stats, which provided a summary of the entire game, we could streamline our dataset without compromising on key metrics like gold accumulation, which we recognized as a vital game-winning determinant.
Our next step was to build a predictive model. Using Random Forests, we not only achieved a prediction accuracy of 91% for game wins but also ranked our features effectively.
Challenges and Breakthroughs 🚀
Despite our success in predictive accuracy, a ranking based purely on the model's scoring system yielded inconsistent results. This realization was a major roadblock, challenging our vision of integrating machine learning into our ranking system.
However, innovation thrives on challenges. We decided to merge our model's score, derived from the ranking of key features, with a team's ELO score. This composite metric gave birth to our final ranking system.
Results and Reflections 🏆
Our final product beautifully aligns with recent tournament outcomes, including the ongoing Worlds Championship. While we are proud of our results, we recognize the dynamism of eSports. Teams that have recently risen in rankings or display a unique playstyle might sometimes be anomalies, reflecting their commendable recent performances.
In retrospect, this project was an enlightening experience. Through trials, errors, and numerous iterations, we've not only crafted a reliable ranking system but also deepened our understanding of data analysis, machine learning, and the unpredictable nature of competitive eSports.
Log in or sign up for Devpost to join the conversation.