Tendência evolutiva e análise preditiva nas copas do mundo de futebol profissional masculino (1930-2018)

Abstract

In soccer, game statistics are an important tool for determining predictive models of performance or outcome. Knowledge of these models can assist coaches in training and game planning. Thus, this dissertation, organized in two independent studies, analyzed the evolutionary trend of the goal average and estimated the odds ratio for the occurrence of results from the game-related statistical data of the FIFA World Cup of Men's Professional Soccer (CMFIFA). In study 1, all 900 matches of the 21 CMFIFA performed between the years 1930 and 2018 were analyzed. The data were organized in: goals scored and suffered in the first and final games; and partial results (first time) and final results (win, draw, and lose). Descriptive and trend analyzes were carried out to verify the evolution of the goal average in the editions of the CMFIFA studied. A cross table was used to check the final scores of the games, followed by an association analysis between partial (first time) and final results of all matches, and a multinomial logistic regression to identify the odds for final victory from the partial results (first time). The results showed a polynomial trend of the mean of goals (p ≤ 0.01; r = 0.34) with the highest mean for the 1954 CMFIFA Switzerland (5.23 ± 2.86) and the lowest mean for CMFIFA Italy 1990 (2.1 ± 1.56). In 64% of the games the scores ended between 0x0 and 2x1. Teams that finished the first half winning had a victory in at least 70% of the games, and never a team lost the match when the first half result was a victory for ≥2 goals difference. On the other hand, losing in the first half increased by more than 100 times the chance of losing the match (OR = 102.6, p <0.01). In study 2, we analyzed the official game-related statistics of all 384 matches of CMFIFA South Africa 2010, Brazil 2014 and Russia 2018. Five independent variables categorized from the median (effective ball possession, red cards, finalizations (%), finals / goals ratio (%) and first-half results. Soon after, a multinomial regression analysis was performed, with the final results of the matches as a result (victory, draw and lose). The model was able to predict above 70% of the overall results, from 80% in predictions of victory results. The chances of success (win in the match) were higher by 50 times with victory in the first half, 14 times with the draw / goals ratio> 7.7%, and four times if they do not receive red cards. We conclude that the goal average of CMFIFA has an undulatory tendency; there are prevalence of placards ≤3 goals; partial victory is associated with the final victory and defeat in relation to the partial victory increases the chance of defeat at the end of the game; victory over partial defeat, not taking red cards and having a better goals / finish line represent a better chance of winning at the end of the games.