Skip to main content

An Unbiased Player Performance Index

Reducing performance of a player to a single scalar value is perhaps the single most valuable holy grail of the sports data industry. In fact, crafting and communicating Key Performance Indicators (KPIs) constitutes the backbone of efficient measurements for almost all sectors in the market. Football media is no stranger to the concept since its beginnings. Journalists have been publishing their perceived scores of players since decades. With the advent of the football data revolution the possibility of using computer algorithms to attain much more objective and accurate indicators became a reality. Using the abundant raw event data, various sports tech companies have developed their own KPIs based on specific algorithms, however it is not a secret that these numbers do not satisfy the industry; found to be too biased (over-valuing certain types of actions etc.) or inaccurate (being far from the definition of a performant player that a scout has in mind).

Comparisonator was founded on this very basis; being aware how scouts, journalists and agents were in need for a near faultless KPI for player performances. Our data science team invested heavily in the subject. Spending countless hours in years to create the “perfect index”; the master index to satisfy all, scouts, journalists, agents, investors, fans, even players themselves, practically everyone in the football universe. 

Effective Use of Machine Learning and AI Techniques

Our research team composed of expert scholars in the field has leveraged the state-of-the-art machine learning and artificial intelligence (AI) techniques to the full extent during the development process. Today, with abundance of available software and cloud services, many may use machine learning algorithms but generally in a superficial manner, with a basic click, without evaluating all the composing nuts and bolts of the matter. However, it is well known that this approach is doomed to fail, the main reason behind the discontent of the clients of many other providers. As we mentioned, this was our specific founding ethos, to craft an immutable and universal KPI, which we call “Comparisonator AI Index”.

During the development process, we placed AI at the heart of research, but not using it ‘blindly and unresponsibly’. Our experts of machine learning research with academic background were well aware of the possible statistical blind spots of the algorithms. They were also very aware of the epistemological nature of the statistics, to say more clearly : the statistics, therefore machine learning and AI is something almost always very subjective. Any attempts to model a phenomenon will yield unsatisfactory results at some point if they are used blindly. Therefore, the only true approach for practical cases is to go “reverse”; which means, we should start with the needs of the clients, scouts, agents and rest; correctly identify and interpret these needs than build the models.

Result of Experience and Academic Research

Our R&D team has spent more than 5 years perfecting the ‘Comparisonator AI Index’. And we have incorporated numerous scouts in the process since the beginning from all over the globe due to aforementioned reasons. Consulting them repeatedly in any incremental step, this communication cycle helped us to develop the perfect index. Contrary to the general approach, a single machine learning algorithm was not developed in an end-to-end manner. As mentioned, the problems of the sector today stem from this fallacy. In contrast, a set of cutting edge machine learning algorithms and statistical models were for diverse tasks. This decomposition helped us to communicate the intermediate findings to scouts who were assisting the development. In turn, our researchers could have iteratively improved the data filtering and processing. Among utilized techniques, there were state-of-the-art interpretability algorithms, deep learning architectures, statistical uncertainty quantification methods and many more. 

Today, we are proud to see the satisfaction of our clients of the ‘Comparisonator AI Index’. Hundreds of scouts and agents are giving very plausible feedback about the accuracy and fairness of the KPI. It helps them to make correct and actionable decisions. We are proud to see how it is sparking hot debates on social media, where millions of fans discuss the success of our index to rank players. In this short white paper, we wanted to give a glimpse of the ideas and laborious development process behind this ‘single perfect number’ we generate everyday for you and how it became so differential from the competitors in the market. 

Dr Eren Ünlü

Dr Eren UNLU was born in Istanbul, in 1989. He has been conducting various academic research projects in machine learning and artificial intelligence. He has a particular interest on the application of artificial intelligence and new statistical techniques in football.