2.1 Combining the Data

a <- read.csv("england_all.csv") %>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1) %>% mutate(Tier = "top")

b <- read.csv("spain_all.csv") %>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1)%>% mutate(Tier = "top")

c <- read.csv("german_all.csv") %>%
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1)%>% mutate(Tier = "top")

d <- read.csv("italy_all.csv")%>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1)%>% mutate(Tier = "top")

e <- read.csv("d1_all.csv")%>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1)%>% mutate(Tier = "second")

f <- read.csv("other_all.csv")%>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1)%>% mutate(Tier = "second")

g <- read.csv("F1_18.csv") %>% 
  mutate(Season = 2018) %>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1) %>% mutate(Tier = "second")

a <- a[,names(d)]
b <- b[,names(d)]
c <- c[,names(d)]
e <- e[,names(d)]
f <- f[,names(d)]
g <- g[,names(d)]
h <- h[,names(d)]

d <- rbind(a,b,c,d,e,f,g,h)

d <- distinct(d)


d <- d %>% 
  select(Season,Date,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HC,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1,Tier) %>% 
  mutate(Date = as_date(dmy(Date)),
         HTG = FTHG+FTAG-HTHG-HTAG,
         month = month(Date, label = TRUE),
         day = wday(Date, label = TRUE),
         dayofmonth = day(Date),
         tsot = AST+HST,
         fav = ifelse(B365H<B365A,"H","A"),
         winner = ifelse(FTR == fav,"fav","ud")) %>% 
  select(Season,Date,month,day,dayofmonth,HomeTeam,AwayTeam,FTHG,FTAG,FTR,HTHG,HTAG,HTR,HTG,HC,tsot,HS,HST,HF,HY,HR,AC,AS,AST,AF,AY,AR,B365H,B365D,B365A,BbAv.2.5,BbAv.2.5.1,Tier,winner,fav)

head(d)

2.2 Preprocessing

In this preprocess stage we will be processing the moving averages of certain variables in order to correctly portray the form of the current teams. There will be 2 moving averages being calculated here, one for the last 3 games, and 1 for the last 10 games. The goal is for the the 3 game moving average to represent more recent form, and the 10 game moving average to represent a sort of power ratings of the clubs that are playing.

Here we do this to a number of variables:

• ‘ma.hgoals10’ : Moving Average of the Home Team Goals in the last 10 Matches
• ‘ma.agoals10’ : Moving Average of the Away Team Goals in the last 10 Matches
• ‘ma.hgloss10’ : Moving Average of the Home Team Goal Loss in the last 10 Matches
• ‘ma.agloss10’ : Moving Average of the Away Team Goal Loss in the last 10 Matches
• ‘ma.hsot10’ : Moving Average of the Total Number of Shots on Target in the last 10 Matches of Home Team
• ‘ma.asot10’ : Moving Average of the Total Number of Shots on Target in the last 10 Matches of Away Team
• ‘ma.hfouls10’ : Moving Average of the Home Team Fouls in the last 10 Matches
• ‘ma.afouls10’ : Moving Average of the Away Team Fouls in the last 10 Matches
• ‘ma.hyellows10’ : Moving Average of the Home Team Yellow Cards in the last 10 Matches

• ‘ma.ayellows10’ : Moving Average of the Away Team Yellow Cards in the last 10 Matches

• ‘ma.hgoals3’ : Moving Average of the Home Team Goals in the last 3 Matches
• ‘ma.agoals3’ : Moving Average of the Away Team Goals in the last 3 Matches
• ‘ma.hgloss3’ : Moving Average of the Home Team Goal Loss in the last 3 Matches
• ‘ma.agloss3’ : Moving Average of the Away Team Goal Loss in the last 3 Matches
• ‘ma.hsot3’ : Moving Average of the Total Number of Shots on Target in the last 3 Matches of Home Team
• ‘ma.asot3’ : Moving Average of the Total Number of Shots on Target in the last 3 Matches of Away Team
• ‘ma.hfouls3’ : Moving Average of the Home Team Fouls in the last 3 Matches
• ‘ma.afouls3’ : Moving Average of the Away Team Fouls in the last 3 Matches
• ‘ma.hyellows3’ : Moving Average of the Home Team Yellow Cards in the last 3 Matches

• ‘ma.ayellows3’ : Moving Average of the Away Team Yellow Cards in the last 3 Matches

#Moving Average Goals 3

Clubs <- levels(d$HomeTeam)
temp1 <- NULL
temp <- NULL

for(y in 2016:2019){
for(x in Clubs){
  
  temp <- d %>% 
  filter(HomeTeam == x|AwayTeam == x,
         Season == y) %>% 
  arrange(Date) %>% 
  mutate(goals = ifelse(HomeTeam == x,FTHG,FTAG),
         hgoals = ifelse(HomeTeam == x,1,0),
         agoals = ifelse(AwayTeam == x,1,0),
         lag1 = lag(goals)) %>% 
  select(Season,HomeTeam,AwayTeam,Date,goals,hgoals,agoals,lag1)
  
  temp1 <- rbind(temp1,temp)
}
}
movingavg.goals <- temp1 %>%
  mutate(m.avggoals=rollapply(lag1,3,mean,align='right',fill=NA))

movingavg.goals <- movingavg.goals %>% 
  mutate(ma.hgoals = ifelse(hgoals == 1,m.avggoals,0),
         ma.agoals = ifelse(agoals == 1,m.avggoals,0)) %>% 
  group_by(Date,HomeTeam,AwayTeam) %>% 
  summarise(ma.hgoals = sum(ma.hgoals),
            ma.agoals = sum(ma.agoals)) %>% 
  ungroup()


d<- merge(movingavg.goals,d,by=c("Date","HomeTeam","AwayTeam"))

#Moving Average goals 10 

Clubs <- levels(d$HomeTeam)
temp1 <- NULL
temp <- NULL

for(y in 2016:2019){
for(x in Clubs){
  
  temp <- d %>% 
  filter(HomeTeam == x|AwayTeam == x,
         Season == y) %>% 
  arrange(Date) %>% 
  mutate(goals10 = ifelse(HomeTeam == x,FTHG,FTAG),
         hgoals10 = ifelse(HomeTeam == x,1,0),
         agoals10 = ifelse(AwayTeam == x,1,0),
         lag1 = lag(goals10)) %>% 
  select(Season,HomeTeam,AwayTeam,Date,goals10,hgoals10,agoals10,lag1)
  
  temp1 <- rbind(temp1,temp)
}
}
movingavg.goals10 <- temp1 %>%
  mutate(m.avggoals10=rollapply(lag1,10,mean,align='right',fill=NA))

movingavg.goals10 <- movingavg.goals10 %>% 
  mutate(ma.hgoals10 = ifelse(hgoals10 == 1,m.avggoals10,0),
         ma.agoals10 = ifelse(agoals10 == 1,m.avggoals10,0)) %>% 
  group_by(Date,HomeTeam,AwayTeam) %>% 
  summarise(ma.hgoals10 = sum(ma.hgoals10),
            ma.agoals10 = sum(ma.agoals10)) %>% 
  ungroup()


d<- merge(movingavg.goals10,d,by=c("Date","HomeTeam","AwayTeam"))

We Separate the december matches of 2019 as a data test to the model later. We come up with the final database of ‘d’ and our later data test of ‘decembertest’.

#We Separate the december matches of 2019 as a data test to the model later
decembertest <- d %>% 
  filter(Season == 2019,
         month == "Dec")

d <- d %>% 
  group_by(Date) %>% 
  tail(-nrow(decembertest)) %>% 
  ungroup()

d <- na.omit(d)

head(decembertest)

After our pre process stage, we end up with our final dataframe and datatest.

head(decembertest)

head(d)

3.1 Variable Exploration

We try to visualize the data by their variables as to get a picture of the distribution of the data and how it affects the final result.

Here we plot the months against the total goals in a match to get a picture of how the general productivity of the teams are through out the year. As we can see there are certain months that show lower productivity in goals than other months. This i presume can be because of team fixture schedules getting tighter in certain months than others, also the change of the seasons can also affect productivity as changes in general temperature between seasons are common in europe.

# We Plot month to the total number of goals in a match
general_productivity <- d %>% 
  group_by(month) %>% 
  summarise(averagetotalgoals = mean (FTAG+FTHG)) %>% 
   ungroup() %>% 
  mutate(scaled_productivity = scale(averagetotalgoals))


productivity_month_plot <- ggplot(general_productivity,aes(x=month, y=averagetotalgoals))+
  geom_col(aes(fill = averagetotalgoals))+
  theme_minimal()+
  scale_x_discrete()+
  labs(title = "Productivity in Matches across the Year",
       y = "Average Total Goals in a Match")+
  theme_minimal()

productivity_month_plot

In this graph we plot our own moving average home goals variable (ma.hgoals10) against the full time goals of the homeside (FTHG) in order to see the distribution, and get a picture on how it affects the final home goal result. As we can see, there is a positive trend, in which the higher the ma.hgoals10 is the more likely home goals will be produced.

#to add the total number of goals to the dataframe
d$totalgoals <- d$FTHG+d$FTAG

# We Plot the moving average goals of the last 10 matches of the home team and how it distributes to the average number of goals in a match
FTHG_ma.hgoals_plot <- d %>% 
  group_by(FTHG) %>% 
  summarise(average.mahgoals10 = mean(ma.hgoals10)) %>% 
   ungroup() %>% 
  mutate(scaled_productivity = scale(average.mahgoals10))

ggplot(FTHG_ma.hgoals_plot,aes(x=FTHG, y=(average.mahgoals10)))+
  geom_col(aes(fill = average.mahgoals10))+
  theme_minimal()+
  scale_x_continuous(breaks=c(1:10))+
  labs(title = "How the Moving Average of Goals from the Home Team effect Home Productivity",
       y = "Average Goals of the Home Team in a match",
       x = "Full Time Home Goals (FTHG)")+
  theme_minimal()

In this third graph we plot the average days of teams have in between league fixtures and see how it affects the home team goal productivity (FTHG). Here we can see a rather different distribution from our second graph, in which here it seems that there is more of an optimum value where if hometeams have more than 5 days off the resting time becomes counterproductive and yields poorer productivity on the pitch.

# We Plot the total days off of the home team and how it effects to the average number of goals in a match
FTG_daysoff_plot <- d %>% 
  group_by(hdaysoff) %>% 
  summarise(average.goals = mean(FTHG+FTAG)) %>% 
   ungroup() %>% 
  mutate(scaled_productivity = scale(average.goals)) %>% 
  head(13)

ggplot(FTG_daysoff_plot,aes(x=hdaysoff, y=(average.goals)))+
  geom_col(aes(fill = average.goals))+
  theme_minimal()+
  scale_x_continuous(breaks=c(1:14))+
  labs(title = "How the Number of days off of the Home Team effect Home Productivity",
       y = "Full Time Home Goals (FTHG)",
       x = "Average Days Off of the Home Team before the match")+
  theme_minimal()

Further visualization can be done, but what we can take from these visualizations are that various variables from the dataframe effect results of the match result, some have different distributions than others, for example average daysoff, has a more optimum value that is diffent from the ma.hgoals10 variable where the higher the value the higher the outcome.