FIFA club world cup 2025

Forecasting game results using Bayesian regression models

Location: bets/vignettes/38.fifa_club_world_cup_25

https://www.transfermarkt.com/fifa-klub-wm/gesamtspielplan/pokalwettbewerb/KLUB/saison_id/2024

rm(list = ls())
set.seed(999)
invisible(sapply(c('tidyverse', "brms", "readr", 
                   "janitor", "lubridate", "stringi", 
                   "DT"), 
                 require, character.only = TRUE))

options(dplyr.summarise.inform = FALSE)

# load functions 
invisible(sapply(
    list.files(pattern = "[.]R$", 
               path = "~/Documents/bets/R", 
               full.names = TRUE), source))

1 Introduction

The tournament is played in five rounds (R), the 32 participating teams are grouped into 8 groups, each group consisting of four teams that compete in a round-robin, the first two of each group qualify for the round of 16, the winners of the round of 16 matches qualify for the quarter-finals, the winners of the quarter-finals qualify for the semi-finals and finally the championship game is decided between the winners of the semi-finals.

2 Round dates

• R1 Round-robin: Jun 14, 2025 to Jun 27, 2025
• R2 Round 16: Jun 28, 2025 to Jul 2, 2025
  
• R3 Quarter-final: Jul 4, 2025 to Jul 5, 2025
  
• R4 Semi-finals: Jul 8, 2025 to Jul 9, 2025
  
• R5 Championship game: Jul 13, 2025

3 Work flow

setwd("~/Documents/bets/data")
flow <- read_csv("fifa_world_cup25_sch.csv")

PrintDT(flow, 7, "Work flow", 0)

4 Data for model

date, round, group, home_team, home_goal, away_goal & away_team

setwd("~/Documents/bets/data")
dm <- read_csv("fifa_world_cup25.csv", col_names = TRUE) %>%
  mutate(home_team = stri_trans_general(
    home_team, id = "Latin-ASCII"),
    away_team = stri_trans_general(
      away_team, id = "Latin-ASCII")) %>%
  mutate(home_goal = as.numeric(home_goal), 
         away_goal = as.numeric(away_goal),
         home_gd = home_goal - away_goal,
         away_gd = away_goal - home_goal,
         date = mdy(date)) %>% 
  select(date, round, group, home_team, away_team, home_goal, 
         away_goal, home_gd, away_gd) %>%
  arrange(date, home_team)

PrintDT(dm, 4, "Input data", 1)

In the round-robin phase, each team plays its matches in three sub-rounds. After each sub-round ends, a Bayesian model is calculated and used to “predict” the results of the following sub-rounds. This process is repeated for each sub-round or round until the tournament ends.

5 Round to be analized

• DM3: Data for model
  
• M3: Bayes model
• DP3: Data used to predict outcomes
  
• PR3: Prediction results
• MQ3: Model quality

5.1 DM3: Data for model

Select data from games that have already finished

# data for model 
dm3 <- dm %>% filter(round == "r11" | round == "r12" | round == "r13" | round == "r2")

5.2 M3: Bayes model

The Bayesian model uses data from finished games and allows the results of future games to be predicted.

5.3 DP3: Data used to predict outcomes

Define the games you want to be predicted, generally referring to matches that have not yet taken place.

# data for prediction
dp3 <- dm %>% 
  filter(round == "r3") %>%
  select(date, group, home_team, away_team, 
         hg = home_goal, ag = away_goal)

5.4 PR3: Prediction results

The Bayesian model estimates the probabilities (winning, drawing or losing the match) and additionally calculates the number of expected goals for each team.

# Using "allow_new_levels" = TRUE  
# Include games after certain dates 
# run predict_matches function 
pr3 <- predict_matches(dm, dp3, m3_bmh, m3_bma)

# model results
pr3a <- model_results(pr3, dp3)

# add group to model results 
pr3b <- full_join(dp3, pr3a, by = join_by(date, group, home_team, away_team, hg, ag))

PrintDT(pr3b, 20, "R1: model results", 2)

print(pr3b)

5.5 MQ3: Model quality

The quality of the model essentially compares the expected results vs. the actual results of the matches.

# model quality
pr3b %>% 
  group_by(model_quality) %>% 
  summarise(freq = n()) %>%
  ungroup() %>%
  mutate("model quality %" = round(100*freq/sum(freq), 1))

pr3b %>% mutate(ph_pa = ph - pa)

6 Tournament results

# league_results
PrintDT(league_results(
  dm %>% select(home_team, away_team, 
                home_score = home_goal, 
                away_score = away_goal)), 
  4, "Tournament results", 1)

teams <- c("Real Madrid", "Bayern Munich", "Paris SG",
           "Fluminense", "Palmeiras", "Chelsea",
           "Bor. Dortmund", "Al-Hilal")

ht_hg_ag <- dm %>% 
  group_by(home_team) %>% 
  drop_na(home_goal) %>%
  summarise(ng_ht = n(), 
            hg_ht = mean(home_goal, na.rm = T), 
            ag_ht = mean(away_goal, na.rm = T)) %>% 
  ungroup() %>%
  filter(home_team %in% teams)

at_hg_ag  <- dm %>% 
  group_by(away_team) %>% 
  drop_na(home_goal) %>%
  summarise(ng_at = n(),
            hg_at = mean(home_goal, na.rm = T), 
            ag_at = mean(away_goal, na.rm = T))  %>% 
  ungroup() %>%
  filter(away_team %in% teams)

da <- cbind(ht_hg_ag, at_hg_ag) %>%
  mutate_if(is.numeric, round, 1) %>%
  mutate(sht = hg_ht - ag_ht,
         sat = ag_at - hg_at)

da1 <- full_join(pr3b, da, 
                 by = join_by(home_team, away_team)) %>% 
  select(home_team:away_team, ph:wbyp, ng_ht:sat)