Dr. Hab., Associate Professor A.A. Polozov¹
Decorated Master of Sports K.B. Tszu²
PhD E.G. Shurmanov^1
1 Ural Federal University, Yekaterinburg
²Academy of Boxing named after K.B. Tszu, Yekaterinburg

Keywords: boxing, rating, information technologies, neural network.

Background. For the last decades the Russian sports community and government have taken many projects to encourage progress of the national boxing sport – albeit their benefits are still contradictory. We feel that popularity of boxing in the country is likely to fall in the near future. To keep up the popularity at a reasonable level, it could be beneficial to offer a fair progress rating system with incentives and fair test toolkits and thereby lay an informational foundation for objective ratings in the national boxing sport. Lately we piloted a neural network project to automate the refereeing service. A key element of the neural network is a small sensor fixed in the boxer's gloves to read the punch-specific pressure peaks.

Objective of the study was to analyze benefits of modern automatic rating systems for boxers and referees.

Methods and structure of the study. Let us compare the most popular rating systems applied in boxing nowadays. Boxing League of Russia rusboxing.com [1] runs the following rating system: 50 - technical draw, 100 – win on scores; 130 – win by technical knockout; 100 – win with disqualification of the opponent; 100 – win due to the opponent’s absence; and 20 - loss.

BoxRec.com [2] is the highly popular boxing site with detailed fight statistics run its own rating system. When a boxer with r_a rating defeats a boxer with r_b rating, with the win value v and pure decision factor cd, the new ratings new_a and new_b upon the fight (earn_f is 33.3%) are calculated as follows

* earn= earn_f * v * ($r_b*cd + (r_b-r_a)/(1+2*cd));

* new_a= r_a + earn;
* new_b= r_b – earn;

Example: Boxer A knocks out Boxer B and they are awarded 1000 and 500 points, respectively, with v=1, cd=1

* earn= 0.33 * 1 * (500*1 + (500-1000)/(1+2*1)) = 111
* r_a_new = 1000 + 111 = 1111               * r_b_new = 500 – 111 = 389

Fightnews.ru online boxing magazine [3] applies the traditional Elo method, with the Boxer A chances to defeat Boxer B calculated as follows.

 EA = 1 / (1 + 10 ** ((RB-RA) / 400)), where:

** Degree of the value; EA - expected score of Boxer A in the fight; RA – Boxer A rating; and RB - Boxer B rating. New rating of Boxer A is calculated as follows:

   R1A = RA + K * (SA - EA) * Kt, where:

 K is the factor that equals 200 for ratings above 2400; 300 for ratings under 2400; and 500 for newcomers;  SA - actual score of Boxer A (1 point for win, 0.5 for a draw and 0 for defeat).

Of special interest is the new WBO rating model offered by the expert community [4] to factor in the past wins and defeats, rank of the tournament, individual qualification statistics, popularity on TV, social contacts, drug-free records etc. And the Boxing Federation of Russia runs its own rating of boxers, but avoids publicity of the rating formula – and this fact may be interpreted as indicative of its dubious quality. We can only presume that the Federation uses some version of the BoxRec.com model.

On the whole, the existing rating systems in boxing: (1) Typically offer immature (childish) solutions for mature problems; (2) Are applied only to elite boxers and turn a blind eye to at least half of the competitors; (3) Give little if any correlations of the actual rating with the real competitive accomplishments; and (4) Keep independent unbiased parties away from the rating system control.

Results and discussion. Our analysis of the existing rating systems showed that their progress is still restricted by the applied information technologies that ideally should:

• Keep accurate statistics of the boxing classes;

• Keep records of the boxing elite expected to win the top-ranking tournaments;

• Report progresses of the regional Boxing Federations to predict competitive accomplishments of the regional boxing elites in 5-15 years;

• Select and encourage progresses of the most gifted boxers;

• Nominate the best coaches based on competitive successes of their trainees;

• Rate benefits of the applied rating methodology;

• Report returns on investments (ROI) in the regional Boxing Federations; and

• Keep individual rating of every boxer even when the fight statistics are not publicized.

We offer a new rating algorithm [5] that works as follows. It is a common knowledge that an individual rating reports two logistic curves - ascend and descend. The individual statistics accumulated by the rating system will make it possible to predict the individual progress peaks and select the best prospects. Variations of the indicators across different coaching schools will make it possible to keep ratings of the coaches and analyze their coaching styles, achievements and drawbacks. Such rating system supported by a cutting-edge application information technologies will be advantageous for the national boxing schools. The rating system should be established on an independent platform of a Big Data type to collect the key data, process and produce highly-dependable ratings, with the actual competitive accomplishments well matched with the grounded expectations.

The refereeing service subjectivity is still ranked among the key problems of the national boxing sport. Referees often avoid reporting the punching statistics simply for the only reason that they do not count the punches in fact. Even when the landed punches are deemed scored by the referees, approximately 30% of the boxing matches end up in team disagreements with the verdicts and protests. It has become quite traditional for the national mass media to discuss rather the refereeing service quality than the fight itself. To put the refereeing service on an unbiased basis, we launched an neural-network-driven automated refereeing system that applies five video cameras: four in the corners and one on the referee’s head. The system neural network will decide whether or not a punch should be scored based on the punching power metering by special sensors (as big as a 10-Ruble coin) fixed inside the both boxing gloves.

The global industry offers a good choice of miniature pressure/ punch sensors that may be connected to the neural network to fix the punch-specific palm pressure. The neural network rated for more than 1-5 million snapshots will sort out non-scoring punches to gloves, elbows, forearms and score target power punches. One more neural network will individualize the fighters and referees and keep the punching and refereeing service statistics. At this juncture the new rating system is tested by a pilot project. It is not improbable that the system will be accessible on a cloud server in Moscow to process the fight video records incoming from all over the country. It will equip the Boxing Federation of the Russian Federation with a new boxing sport control and management toolkit. The solutions offered by the system will put the match refereeing service on a fair basis as it will secure flows of scoring power punches rather than total unclassified punches as is the case at present.

Conclusion. It is recommended for the national boxing community to give a special priority to the modern information technologies for Big Data flow applied by a well-designed rating system to secure good matching of the actual competitive accomplishments with the grounded expectations; with the refereeing service put on an objective basis by using modern neural network that will classify punches by power and accuracy and sort out non-scoring punches (in contrast to the existing 100% registration of punches) being fully insensitive to pressure from the supporters and vested interests.

References

1. rusboxing.com
2. BoxRec.com
3. fightnews.ru
4. WBO.com
5. Karminskiy A.M., Polozov A.A. Encyclopedia of rating: economics, sport, society. Moscow: Ekonomika i zhizn publ.. 2011. 455 p.

Corresponding author: d_narkhov@mail.ru

Abstract

Objective of the study was to analyze benefits of modern automatic rating systems for boxers and referees..
Methods and structure of the study. To develop the author's rating computation algorithm in boxing, we made a comparative assessment of the approaches used.

Study results and conclusions. The best way to implement the proposed rating computation algorithm is to accumulate data over time by the ascension and degradation periods in an athletic career, which will enable us to determine the highest skill level and allocate the most talented boxers. The analysis of changes in this indicator when working with different coaches will help rank coaches and their training methodologies. The establishment of an advanced data environment creates an advantage for the national boxing school.

The automated refereeing project based on the developed neural network suggests that, instead of judges, a bout will be watched through 3 video cameras: 4 in the corners of the ring and 1 on the body of a referee. A specially trained neural network will decide whether to count a strike or not. It will be activated at the moment of impact. To understand when this moment arrives, a special sensor should be chosen and given to a boxer to hold in both hands. The solution to this problem will help decide the winner not by the number of strikes but by the total strike weight.
The authors conclude that the employment of informational resources to create Dig Data in boxing with the most adequate rating computation for comparing the expected and actual results; the use of neural networks to solve the current problem in boxing - refereeing - are the priority areas in the boxing development process.