What is the percentage of coincidence of the forecast with the real result?

We tell about the research of our graduate student Alexander Krutikov. The main experiment of his scientific activity is forecasting with an accuracy of 60-70% of the outcome of various sports competitions based on neural networks

- Alexander, good afternoon! We congratulate you on your professional holiday! Please tell us why you decided to continue your studies as a postgraduate? And did not complete the educational path, as it happens usually?

- Probably, there are three reasons: 1) It remained an interesting topic for me even since magistracy. And by and large, in the magistracy, I just started researching it. It all started with simple laboratory work in the discipline related to neural networks. The topics of the laboratory works were easy for me, and I decided to apply them in the applied field, with which I myself was actively involved. This is the area of ​​physical culture and sports. I wrote a master's thesis on this topic, but this seemed to me not enough, and it was decided to try my hand at graduate school. 2) An invitation letter was received from the university, in which the idea of ​​further continuing education was voiced. 3) At that time, there was a desire to connect life, including with teaching.

- And now about your scientific brainchild. What led you to experiment with living and artificial intelligence?

- Banal interest and curiosity! I have always been attracted by the topic of forecasting, there is something in it. In sports, forecasts play an important role, it is a shame, however, that for the majority it is only in the betting and gambling context. But for the athletes themselves, this is an important part, as well as for the coaches.

Preparation, training can be built on the basis of the predicted results of this very training, not to mention the forecast for specific events or tournaments.

- Alexander, we know that you are a pro in athletics. We had a large number of Championships of various levels. We cannot but ask: can you predict your results? Can predictions based on neural networks at all increase the odds of winning?

In fact, I am a representative of the disciplines of athletics throwing. And I conducted my first experiments on my own results. To be honest, I got bored with it quickly, but certain predictions came out. On the question of chances, it is possible to change training preparation, and, therefore, to some extent, the chances of a favorable outcome for the athlete.

All forecasts are made on a training sample, which is based on existing data. Depending on what we predict, the available data may be individual test results of an athlete, team statistics, statistics of matches, and, accordingly, their participants, medal statistics of countries, etc.

When compiling a training sample, the opinion of an expert on the kind of sport about which the forecast is made plays an important role. It is necessary to find out which parameters to include in the selection. For example, two of the most interesting predictions that I made are the NKZ forecast for the 2019 World Universiade and the 2020 Olympic Games that have not yet taken place (2021). In the case of the Universiade, the winner of the NKZ was unambiguously predicted, and also some of the leaders of the medal standings. The OI forecast has not yet been verified (this year it will be updated).

- What other possibilities does the study of neural networks provide? Can it be realistic also to predict human behavior in certain situations?

There are enough opportunities. Starting from the already banal image recognition, to writing masterpieces (or almost masterpieces) of music. Yes, it is possible to predict human behavior, and there are systems that, in fact, are doing this. Systems that learn from big data may well predict human behavior in social spheres, marketing, advertising, etc.

- It turns out that we can use your research within the walls of the university? Applied order. Can teachers predict the progress of a group in advance, for example?

- I think if we define some control points of the group, fix them for several periods and then draw up a training sample, then in the future it is possible to predict the average performance of the group. It will only work with one specific group. For each new group, you will have to make a new corresponding sample.

- Do you plan to expand your area of ​​research in the future? What can they potentially touch?

- I would like to make similar experiments in economics, meteorology, climatology, astronomy, in the sciences, where a numerical forecast is required. If we talk about the economy, this is, first of all, a forecast of changes in stock quotes, the percentage of annual inflation, and the occurrence of risks. It is interesting to experiment with the forecast time - from ultra-short-term to ultra-long-term.

- What are your plans for defending your dissertation?

It is difficult to make plans regarding protection now, because this is a rather difficult process and much remains to be done and done in stages. I set goals gradually, and the first of them is to write certain sections of the dissertation. With my head, associate professor of the Computer Department Vasily Yurievich Meltsov, we regularly discuss the results that are available at the moment and set new tasks. There is still a lot of work to do.

We wish you brilliant defence, only reliable forecasts and strengthening of your scientific authority. Thank you!

Read in Russian