XI INTERNATIONAL CONFERENCE

“INFORMATION TECHNOLOGY AND IMPLEMENTATION” (IT&I-2024)

An intelligent system for analyzing the dependence of a chess player's rating on inaccurate moves

Andriy Dudnik, Yurii Moroz, Olga Leshchenko, Oleksandr Makhovych and Ihor Kolisnyk

Introduction & Research Question

In chess, the main value that represents the strength of the player is their rating.

But the rating depends on the results of the games, not on the moves that are made on the board.

Elo Rating

Average CentiPawn Loss

Question: How strongly the rating correlates with the moves that players make at different levels?

Methodology

• The games were collected from ChessBase.

​

• The code was written in Python using JupyterLab.

​

• There are 80 players here, divided into 8 rating classes.

​

• Each player has exactly 100 games, excluding short ones.

​

• We analyzed 8000 games using chess engine Stockfish 16.

​

Results

Scatter plot of each player’s ACPL in every game

• The dots represent the ACPL value of every game played by each player.
• The players are sorted by their mean ACPL in descending order.
• Only 4 games out of 8000 ended with more than a 100 ACPL value.

• H & G — Experts
• F — Candidate Masters
• E — FIDE Masters
• D — International Masters
• C & B — Grandmasters
• A — super Grandmasters

Results

Box plot of each player’s ACPL in every game

• The horizontal lines outside the box show minimum and maximum except outliers.
• Half of the data lies inside the box.
• The line inside the box shows the median.
• The dots above are outliers.

• H & G — Experts
• F — Candidate Masters
• E — FIDE Masters
• D — International Masters
• C & B — Grandmasters
• A — super Grandmasters

Results

• blunder — a big mistake
• mistake — the position becomes worse
• inaccuracy — slightly worsens the position
• ok — suboptimal move
• missed win — winning move wasn’t found

​

​

​

• Wesley So is the most precise player.
• The highest-rated player and ex-champion Magnus Carlsen is second.
• Nihal Sarin, who has not even reached a 2700 Elo rating yet, is third.

​

​

​

Wesley made the fewest mistakes and inaccuracies, while Nihal made the fewest blunders (big mistakes).

Players with ACPL value lower than 15

Results

Scatter plot of player’s mean ACPL in all games with a regression line

• no large residuals
• bigger distances in the lower rating classes

​

​

​

The lower rating class, the bigger ACPL range.

• H & G — Experts
• F — Candidate Masters
• E — FIDE Masters
• D — International Masters
• C & B — Grandmasters
• A — super Grandmasters

Results

Correlation heatmap

Rating strongly correlates not only with ACPL but also with all types of mistakes made by the player.

Conclusion & Future Agenda

We compared the quality of play of a wide range of players using ACPL as a main factor.

Also, we added move classifications such as blunder, mistake, and inaccuracy.

​

​

Our results showed a high correlation between rating and ACPL and different types of mistakes.

The correlation between rating and ACPL is -0.95, which is very strong.

​

​

In our future work, we will add some other features, such as a style of play or complexity of the position.

We also plan to build a human-like chess bot based on this research.

Thank you for your attention!