Pith. sign in

REVIEW 2 major objections 1 minor 19 references

League of Legends: Real-Time Result Prediction

T0 review · 2 major / 1 minor · reviewed 2026-05-24 · grok-4.3

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read A LightGBM model predicts League of Legends match outcomes at 81.62% accuracy when 60-80% of match time has elapsed.

desk verdict The 81.62% accuracy claim cannot be evaluated because the data is unpublished and the paper gives no details on splits, features, or validation. read the letter →

arxiv 2309.02449 v1 submitted 2023-09-02 cs.LG

Jailson B. S. Junior , Claudio E. C. Campelo This is my paper
classification cs.LG
keywords LeagueofLegendsreal-timepredictionLightGBMmachinelearningesportsmatchoutcomegameanalytics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper investigates whether machine learning can forecast the winner of a League of Legends match while the game is still underway. It tests several algorithms on variables recorded at successive percentages of elapsed time and relies on unpublished match data. The LightGBM model records the highest accuracy in the 60-80% time window, while logistic regression and gradient boosting perform better in the opening stages. The work positions these predictions as potentially useful for in-game strategy adjustments and for live betting applications. Results are presented as average accuracy figures across different match stages.

What carries the argument

LightGBM classifier trained on game-state features recorded at successive percentages of elapsed match time.

What would settle it

Evaluating the trained LightGBM model on a new collection of matches from a later tournament and measuring whether accuracy remains near 81.62% in the 60-80% elapsed-time window.

Watch

Extended reading notes

Core claim

The paper claims that machine learning models applied to match-state variables can predict real-time outcomes in League of Legends, with a LightGBM model achieving an average accuracy of 81.62% specifically in intermediate stages when 60% to 80% of the match time has elapsed, and with logistic regression and gradient boosting showing stronger results in early stages.

Load-bearing premise

The unpublished match data used for training and testing is representative of future matches and contains no selection bias or information leakage from the outcome.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The paper evaluates several machine learning models (including LightGBM, Logistic Regression, and Gradient Boosting) for real-time prediction of League of Legends match outcomes at different elapsed-time percentages, using an unpublished dataset of matches. It reports that LightGBM achieves the highest average accuracy of 81.62% for matches in the 60-80% elapsed-time window, with other models performing better in early stages, and positions the work as relevant to esports strategy and betting.

Significance. If the performance numbers can be independently verified, the temporal breakdown of accuracy across match stages would be a useful applied result in esports analytics. The use of unpublished data is presented as a strength, but the absence of any public release, feature list, or validation protocol prevents assessment of whether the headline accuracy reflects genuine predictive power or artifacts of data construction.

major comments (2)
  1. [Abstract] Abstract: the central claim of 81.62% average accuracy for LightGBM (60-80% elapsed time) is presented with no accompanying information on number of matches, number of features, train-test split procedure, cross-validation method, or handling of class imbalance, so the numerical result cannot be evaluated for information leakage or selection bias.
  2. [Methods] Methods (data collection and preprocessing sections): no description is given of how per-snapshot feature vectors are constructed at a given time percentage, whether any post-outcome statistics are included, or how the train/test split is performed to guarantee that all inputs at time t are strictly pre-outcome; this directly undermines the verifiability of the reported accuracies.
minor comments (1)
  1. [Abstract] The abstract and results sections would benefit from explicit statements of the total number of matches and the exact feature set used.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for highlighting the need for greater methodological transparency to allow independent verification of the reported results. We agree that the current manuscript lacks sufficient detail on data construction and evaluation procedures, and we will revise accordingly to address these points.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of 81.62% average accuracy for LightGBM (60-80% elapsed time) is presented with no accompanying information on number of matches, number of features, train-test split procedure, cross-validation method, or handling of class imbalance, so the numerical result cannot be evaluated for information leakage or selection bias.

    Authors: We agree that the abstract omits these critical details. In the revised version we will expand the abstract to report the number of matches in the dataset, the number of features, the train-test split procedure (performed at the match level), the cross-validation method, and the approach to class imbalance. This will allow readers to assess the headline accuracy for potential leakage or bias without needing to consult the full text. revision: yes

  2. Referee: [Methods] Methods (data collection and preprocessing sections): no description is given of how per-snapshot feature vectors are constructed at a given time percentage, whether any post-outcome statistics are included, or how the train/test split is performed to guarantee that all inputs at time t are strictly pre-outcome; this directly undermines the verifiability of the reported accuracies.

    Authors: We acknowledge that the Methods section does not explicitly describe snapshot construction, exclusion of post-outcome statistics, or the precise train/test split logic. We will revise the data collection and preprocessing subsections to specify: (i) the exact procedure for building per-snapshot feature vectors at each elapsed-time percentage, (ii) confirmation that only in-game, pre-outcome statistics are used, and (iii) the match-level train/test partitioning that ensures all inputs at time t are strictly pre-outcome. These additions will directly support verifiability of the accuracies. revision: yes

standing simulated objections not resolved
  • We are unable to release the raw unpublished dataset for independent verification due to data ownership and confidentiality constraints, although a complete feature list and validation protocol can be added to the revised manuscript.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical ML accuracies on collected data

full rationale

The paper reports standard supervised learning results: models (LightGBM, logistic regression, etc.) are trained on match snapshots and evaluated for accuracy at different time percentages. No derivation chain, equations, or first-principles claims exist that could reduce to inputs by construction. Performance numbers are direct outputs of fitting and testing on the dataset; they are not renamed predictions or self-referential. No self-citations, uniqueness theorems, or ansatzes are invoked. The result is an ordinary empirical benchmark on private data and does not meet any of the enumerated circularity patterns.

Assumptions & free parameters 1 free parameters · 1 assumptions · 0 invented entities

The paper rests on standard supervised learning assumptions plus the domain claim that real-time game features are predictive; no new entities are postulated.

free parameters (1)
  • LightGBM hyperparameters
    Hyperparameters were chosen or tuned to produce the reported 81.62% accuracy figure.
assumptions (1)
  • domain assumption Match data points are independent and identically distributed
    Implicit in any supervised learning evaluation on game telemetry.

how reviews work

0 comments
Cite this review

Pith. "Pith review of League of Legends: Real-Time Result Prediction." pith.science (2026). https://pith.science/paper/2309.02449

@misc{pith2026230902449,
  author       = {Pith},
  title        = {Pith review of: League of Legends: Real-Time Result Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2309.02449}},
  note         = {Machine review of arXiv:2309.02449}
}
read the original abstract

This paper presents a study on the prediction of outcomes in matches of the electronic game League of Legends (LoL) using machine learning techniques. With the aim of exploring the ability to predict real-time results, considering different variables and stages of the match, we highlight the use of unpublished data as a fundamental part of this process. With the increasing popularity of LoL and the emergence of tournaments, betting related to the game has also emerged, making the investigation in this area even more relevant. A variety of models were evaluated and the results were encouraging. A model based on LightGBM showed the best performance, achieving an average accuracy of 81.62\% in intermediate stages of the match when the percentage of elapsed time was between 60\% and 80\%. On the other hand, the Logistic Regression and Gradient Boosting models proved to be more effective in early stages of the game, with promising results. This study contributes to the field of machine learning applied to electronic games, providing valuable insights into real-time prediction in League of Legends. The results obtained may be relevant for both players seeking to improve their strategies and the betting industry related to the game.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

  1. [1]

    ActivePlayer - League of Legends

    “ActivePlayer - League of Legends.” https://activeplayer.io/ league-of- legends/. Accessed on: May 16, 2023

    work page 2023

  2. [2]

    The Evolution of eSports: An Analysis of its origin and a look at its prospective future growth as enhanced by Information Technology Management tools

    A. H. Olsen, “The evolution of esports: An analysis of its origin and a look at its prospective future growth as enhanced by information technology management tools,” arXiv preprint arXiv:1509.08795, 2015

    work page Pith review arXiv 2015

  3. [3]

    esports: a new industry,

    S. Block and F. Haack, “esports: a new industry,” in SHS Web of Conferences, vol. 92, p. 04002, EDP Sciences, 2021

    work page 2021

  4. [4]

    Campeonato Mundial de League of Legends

    Wikipedia, “Campeonato Mundial de League of Legends.” https://pt. wikipedia.org/wiki/Campeonato Mundial de League of Legends. Ac- cessed on: May 16, 2023

    work page 2023

  5. [5]

    Worlds 2022: Mundial de LoL prom- stages of the game (elapsed time percentage of 20%), other ete ser u´ nico

    Showmetech, “Worlds 2022: Mundial de LoL prom- stages of the game (elapsed time percentage of 20%), other ete ser u´ nico .” https://ww w.showmetech.com.br/ models such as Logistic Regression and Gradient Boosting had superior results. This indicates that different models may be more suitable depending on the stage of the match. Considering an accuracy of...

    work page 2022

  6. [6]

    Dota 2 match prediction through deep learning team fight models,

    C. H. Ke, H. Deng, C. Xu, J. Li, X. Gu, B. Yadamsuren, D. Klabjan, R. Sifa, A. Drachen, and S. Demediuk, “Dota 2 match prediction through deep learning team fight models,” 2022 IEEE Conference on Games (CoG), 2022

    work page 2022

  7. [7]

    Machine learning models for dota 2 outcomes prediction,

    K. Akhmedov and A. Phan, “Machine learning models for dota 2 outcomes prediction,” 2021

    work page 2021

  8. [8]

    Predicting the outcome of online heroes of the storm matches,

    J. Swidler, “Predicting the outcome of online heroes of the storm matches,” 2017

    work page 2017

Show all 19 references
  1. [9]

    Hotslogs

    HOTSLogs, “Hotslogs.” Available at: https://www.hotslogs.com/default. Accessed on: June 8, 2023

    2023

  2. [10]

    Quantitative analysis for win/loss prediction of ’league of legends’ utilizing the deep neural network system through big data,

    S.-J. No, Y.-J. Moon, and Y.-H. Hwang, “Quantitative analysis for win/loss prediction of ’league of legends’ utilizing the deep neural network system through big data,” Journal of the Korea Society of Computer and Information, 2021

    2021

  3. [11]

    League of legends match outcome prediction,

    L. Lin, “League of legends match outcome prediction,” 2016

    2016

  4. [12]

    League of legends: An application of classification algorithms to verify the prediction importance of main in -game variables,

    A. C. S. Cruz, T. G. D. Reˆgo, T. D. M. Filho, and Y. Malheiros, “League of legends: An application of classification algorithms to verify the prediction importance of main in -game variables,” Anais Estendidos do XX Sim po´ sio Brasilei ro de Games e Entretenimento Digital (S...

    2021

  5. [13]

    E- sports player performance metrics for predicting the outcome of league of legends matches considering player roles,

    F. Bahrololloomi, F. Klon owski, S. Saue r, R. Horst, and R. Do¨ rner, “E- sports player performance metrics for predicting the outcome of league of legends matches considering player roles,” SN Computer Science , 2023

    2023

  6. [14]

    Continuous outcome prediction of league of legends competitive matches using recurrent neural networks,

    A. L. C. Silva, G. Pappa, and L. Chaimowicz, “Continuous outcome prediction of league of legends competitive matches using recurrent neural networks,” 2018

    2018

  7. [15]

    Using machine learning to predict game outcomes based on player-champion experience in league of legends,

    T. D. Do, S. I. Wang, D. S. Yu, M. G. McMillian, R. P. McMahan, and R. P. McMahan, “Using machine learning to predict game outcomes based on player-champion experience in league of legends,” arXiv: Learning, 2021

    2021

  8. [16]

    League of legends: A study of early game impact,

    R. D. Gaina, “League of legends: A study of early game impact,” 2018

    2018

  9. [17]

    Official riot games api

    “Official riot games api.” https://developer.riotgames.com/. Accessed on: April 20, 2023

    2023

  10. [18]

    Riot-watcher

    pseudonym117, “Riot-watcher.” https://github.com/pseudonym117/ Riot-Watcher. Accessed on: April 20, 2023

    2023

  11. [19]

    League of Legends Match Data at Various Time Intervals

    J. B. da Silva Junior and C. Campelo, “League of Legends Match Data at Various Time Intervals.” https://doi.org/10.5281/zenodo.8303397, Aug. 2023

    2023 doi

Pith tools

Reviewed May 24, 2026 · model on record in the stance chip above.