Electronic Sports (eSports) are one of the most prevalent entertainments in modern society. As eSports is gaining more and more popularity over time, a proper evaluation system that can accurately analyze and predict players’ performance is indispensable. Therefore, machine learning is proposed as an effective prediction tool in the studying of data from eSports matches so that eSports teams can utilize it to ameliorate their future accomplishments. Two popular eSports game types (Multiplayer Online Battle Arena and First Person Shooter) are analyzed. Common features they share include communication, equipment, economy, and so on. Correct hero pick is the most predominant factor in MOBA games. Research shows that machine learning models like Logistic Regression, Support Vector Machine, and Neural Networks can effectively combine hero selection data to predict the outcome of a game. On the other hand, the most important factor in FPS games is the in-game equipment each round. Additional research proves that Logistic Regression, K-Nearest Neighbors, Random Forest, and Neural Networks are the most powerful models for predicting the team win rate. This paper can serve as an instruction for applying machine learning methods to making predictions in eSports games. Meanwhile, it can provide available techniques to help professional eSports teams improve their competitive level and future in-game understanding as well.