Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predicting Outcomes in Video Games with Long Short Term Memory Networks

Kittimate Chulajata, Sean Wu, Fabien Scalzo, Eun Sang Cha

TL;DR

The paper tackles real-time outcome forecasting in two-player video games by using a health-bar time-series signal as input to LSTM and Transformer classifiers, evaluated at 25%, 75%, and 95% match progression in Super Street Fighter II Turbo. It benchmarks against baseline classifiers (KNN, SVM, RF) and demonstrates that LSTMs achieve competitive or superior ROC-AUC while maintaining feasible inference times, with Transformers showing tendencies toward overfitting on the limited dataset. An open-source dataset and code accompany the study, enabling replication and extension to other games. Overall, the work advances audience-engagement tooling for esports by enabling mid-round predictions and highlights directions such as data augmentation and hybrid architectures for broader applicability.

Abstract

Forecasting winners in E-sports with real-time analytics has the potential to further engage audiences watching major tournament events. However, making such real-time predictions is challenging due to unpredictable variables within the game involving diverse player strategies and decision-making. Our work attempts to enhance audience engagement within video game tournaments by introducing a real-time method of predicting wins. Our Long Short Term Memory Network (LSTMs) based approach enables efficient predictions of win-lose outcomes by only using the health indicator of each player as a time series. As a proof of concept, we evaluate our model's performance within a classic, two-player arcade game, Super Street Fighter II Turbo. We also benchmark our method against state of the art methods for time series forecasting; i.e. Transformer models found in large language models (LLMs). Finally, we open-source our data set and code in hopes of furthering work in predictive analysis for arcade games.

Predicting Outcomes in Video Games with Long Short Term Memory Networks

TL;DR

The paper tackles real-time outcome forecasting in two-player video games by using a health-bar time-series signal as input to LSTM and Transformer classifiers, evaluated at 25%, 75%, and 95% match progression in Super Street Fighter II Turbo. It benchmarks against baseline classifiers (KNN, SVM, RF) and demonstrates that LSTMs achieve competitive or superior ROC-AUC while maintaining feasible inference times, with Transformers showing tendencies toward overfitting on the limited dataset. An open-source dataset and code accompany the study, enabling replication and extension to other games. Overall, the work advances audience-engagement tooling for esports by enabling mid-round predictions and highlights directions such as data augmentation and hybrid architectures for broader applicability.

Abstract

Forecasting winners in E-sports with real-time analytics has the potential to further engage audiences watching major tournament events. However, making such real-time predictions is challenging due to unpredictable variables within the game involving diverse player strategies and decision-making. Our work attempts to enhance audience engagement within video game tournaments by introducing a real-time method of predicting wins. Our Long Short Term Memory Network (LSTMs) based approach enables efficient predictions of win-lose outcomes by only using the health indicator of each player as a time series. As a proof of concept, we evaluate our model's performance within a classic, two-player arcade game, Super Street Fighter II Turbo. We also benchmark our method against state of the art methods for time series forecasting; i.e. Transformer models found in large language models (LLMs). Finally, we open-source our data set and code in hopes of furthering work in predictive analysis for arcade games.
Paper Structure (18 sections, 1 equation, 2 figures, 2 tables)

This paper contains 18 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Existing work
  3. Methodology
  4. Assumptions
  5. Data Collection
  6. Splitting Data into Rounds
  7. Padding Data
  8. Models
  9. Baseline Machine Learning Models
  10. K-Nearest Neighbor Classifier (KNNs)
  11. Support Vector Classifier (SVMs)
  12. Random Forest Classifier (RF)
  13. Long Short-Term Memory Classifier
  14. Transformer Classifier
  15. Training Procedure
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Visualization of ROC Curves comparing LSTM to the Transformer architecture. LSTMs outperform the Transformer to predict round outcomes from 25%, 75%, and also 95%.
  • Figure :