DOTA2 Winner Team Prediction based on Stacked Bidirectional LSTM Network

With the swift development of e-sports technology, games are now using techniques of machine learning in several scopes of application. One of the most significant applications of gaming is gaining insights into match details and various types of predictions. Competitive games provide enormous amounts of data from a match. In the context of data and analytics, the word “insight” refers to the analysis of those game data to find out the pattern or relationship among the variables. In this work, the match stats data as kills-deaths-assists along with the match duration and the heroes has been used to predict the winner of a particular match based on a stacked Bi-LSTM neural network model. The match data which is used in training the model, is collected from the OpenDota API. The results demonstrate the performance of the stacked Bi-LSTM model in predicting the winning team of a game compared to other machine learning classifiers.