Combining Dynamic Time Warping, Dimensionality Reduction and Gaussian Mixtures to Calculate Live Win Probabilities in League of Legends
Part of Qarik's mission is to continuously innovate to ensure success for future partners to go further, faster. The Data Science team is often tasked with digitizing and combining existing data that is lost in the cracks of existing businesses, automating costly manual processes, or creating new products to complement existing offerings.
Recently, Qarik’s Data Science team investigated machine learning models within the context of predicting the outcomes of esports games, focusing on League of Legends (LoL). Experimenting with many different techniques and approaches is necessary when trying to model complicated phenomena like LoL games. Often they fail, sometimes they are promising but don't quite gel with the overall architecture built for the finished approach.
In this case study, we would like to showcase one of the approaches that ended up on the cutting room floor. Dynamic Time Warping (DTW) is a rich framework for comparing different time series, and we show how it can be used in conjunction with two universal data science techniques to predict in real-time the probability of winning a game of LoL.