A comparative analysis of machine learning methods with the application of the Kolmogorov-Gabor polynomial for forecasting sports event outcomes

The article presents a comparative analysis of the effectiveness of machine learning methods for predicting the results of football matches, with a focus on the application of the elementary image of the Kolmogorov-Gabor polynomial. The relevance of the study is due to the need to choose models that are balanced in accuracy, interpretability, and computational complexity in conditions of high stochasticity of sports data. The scientific novelty lies in the adaptation of the elementary image of the Kolmogorov-Gabor polynomial (KGp) for sports analytics tasks and its complex comparison with a wide range of algorithms, from classical regression to gradient boosting. Based on historical data, models have been built and analyzed: an elementary image of a polynomial, linear regression with regularization, a random forest, gradient boosting, and a neural network. The results were evaluated by metrics MAE and accuracy of predicting the outcome. A model based on an elementary image of a polynomial Kolmogorov-Gabor showed competitive accuracy comparable to more complex ensemble methods, while maintaining advantages in computational efficiency and the potential interpretability of the structure of nonlinear dependencies. It was concluded that it is advisable to use this approach as an effective tool for building hybrid forecasting systems in sports analytics.