Time to Die: Death Prediction in Dota 2 using Deep Learning

Abstract

Esports have become major international sports with hundreds of millions of spectators. Esports games generate massive amounts of telemetry data. Using these to predict the outcome of esports matches has received considerable attention, but micro-predictions, which seek to predict events inside a match, is as yet unknown territory. Micro-predictions are however of perennial interest across esports commentators and audience, because they provide the ability to observe events that might otherwise be missed: esports games are highly complex with fast-moving action where the balance of a game can change in the span of seconds, and where events can happen in multiple areas of the playing field at the same time. Such events can happen rapidly, and it is easy for commentators and viewers alike to miss an event and only observe the following impact of events. In Dota 2, a player hero being killed by the opposing team is a key event of interest to commentators and audience. We present a deep learning network with shared weights which provides accurate death predictions within a five-second window. The network is trained on a vast selection of Dota 2 gameplay features and professional/semi-professional level match dataset. Even though death events are rare within a game (1% of the data), the model achieves 0.377 precision with 0.725 recall on test data when prompted to predict which of any of the 10 players of either team will die within 5 seconds. An example of the system applied to a Dota 2 match is presented. This model enables realtime micro-predictions of kills in Dota 2, one of the most played esports titles in the world, giving commentators and viewers time to move their attention to these key events.

Defining the esports bettor: evidence from an online panel survey of emerging adults

“…A unique feature of esports is that it is incredibly data-rich; remarkable volumes of data are outputted from every game (Block et al, 2018). The potential of such data has already been spotted by research groups and commercial organizations for advanced player tracking (Katona et al, 2019) and performance analysis (Schubert et al, 2016). This has potentially important implications for the development of esports betting, which rely on data to generate, regulate, and market products through a variety of media to consumers.…”

Supervised Learning Achieves Human-Level Performance in MOBA Games: A Case Study of Honor of Kings

“…We, thus, summarize representative studies on system-level AI for different types of games, covering Non-RTS, RTS, and MOBA games. Partial Game AI solutions, e.g., predicting match results or time to die for heroes [17], [18], will not be discussed in detail.…”

WARDS: Modelling the Worth of Vision in MOBA’s

“…First, a logistic regression was conducted, which encountered a 73% correlation between the total WARDS for wards that were active at any given instance in the game, and the total gold net-worth of the corresponding team in the following 5 minutes. The difference in a team's total gold net-worth was selected as an overall performance indicator, as it is commonly used for that purpose [10]. It is also a good encapsulation measurement for the purpose of this paper, as it will increase with game events that are related to a ward's performance, such as deaths and lane dominance, as described in previous sections.…”

“…Due to the novelty of the model, particularly its ability to report performance during the running game, there is no consist baseline to be compared. We have looked at similar predicting algorithm, that are aimed at different aspects of the game [10] [8] [9]. Although none of the prediction models have looked at warding, nor a similar time frame of a period of approximately six minutes, we have found the overall performance of the network to be in line with the predictive capabilities we have encountered.…”