Check out the data set - likely columns to drop would be CustomerID, Surname, and we need to create a random ID. Given the broad range of ages, credit scores, balances and salaries, any model would need some kind of normalisation. We’ll do that later after first analysis.
summary(df)
## CustomerId Surname CreditScore Geography
## Min. :15565701 Length:10002 Min. :350.0 Length:10002
## 1st Qu.:15628525 Class :character 1st Qu.:584.0 Class :character
## Median :15690732 Mode :character Median :652.0 Mode :character
## Mean :15690933 Mean :650.6
## 3rd Qu.:15753226 3rd Qu.:718.0
## Max. :15815690 Max. :850.0
##
## Gender Age Tenure Balance
## Length:10002 Min. :18.00 Min. : 0.000 Min. : 0
## Class :character 1st Qu.:32.00 1st Qu.: 3.000 1st Qu.: 0
## Mode :character Median :37.00 Median : 5.000 Median : 97199
## Mean :38.92 Mean : 5.012 Mean : 76491
## 3rd Qu.:44.00 3rd Qu.: 7.000 3rd Qu.:127648
## Max. :92.00 Max. :10.000 Max. :250898
## NA's :1
## NumOfProducts HasCrCard IsActiveMember EstimatedSalary
## Min. :1.00 Min. :0.0000 Min. :0.0000 Min. : 11.58
## 1st Qu.:1.00 1st Qu.:0.0000 1st Qu.:0.0000 1st Qu.: 50983.75
## Median :1.00 Median :1.0000 Median :1.0000 Median :100185.24
## Mean :1.53 Mean :0.7055 Mean :0.5149 Mean :100083.33
## 3rd Qu.:2.00 3rd Qu.:1.0000 3rd Qu.:1.0000 3rd Qu.:149383.65
## Max. :4.00 Max. :1.0000 Max. :1.0000 Max. :199992.48
## NA's :1 NA's :1
## Exited
## Min. :0.0000
## 1st Qu.:0.0000
## Median :0.0000
## Mean :0.2038
## 3rd Qu.:0.0000
## Max. :1.0000
##
Checking for missing data. There seems to be a very small number of nulls or empty values.
count_missing <- function(data) {
data %>% summarise_all(~ sum(is.na(.)))
}
missing_counts <- count_missing(df)
print(missing_counts)
## # A tibble: 1 × 13
## CustomerId Surname Credi…¹ Geogr…² Gender Age Tenure Balance NumOf…³ HasCr…⁴
## <int> <int> <int> <int> <int> <int> <int> <int> <int> <int>
## 1 0 0 0 1 0 1 0 0 0 1
## # … with 3 more variables: IsActiveMember <int>, EstimatedSalary <int>,
## # Exited <int>, and abbreviated variable names ¹CreditScore, ²Geography,
## # ³NumOfProducts, ⁴HasCrCard
Check the total splits across the data set for different variables. There’s a better way of doing and displaying this, but didn’t want to waste time here.
table_percentage <- function(column) {
prop.table(table(column))
}
table_percentage <- function(column) {
prop.table(table(column)) * 100
}
print(table_percentage(df$Exited))
## column
## 0 1
## 79.62408 20.37592
print(table_percentage(df$Geography))
## column
## France Germany Spain
## 50.13499 25.09749 24.76752
print(table_percentage(df$Gender))
## column
## Female Male
## 45.43091 54.56909
print(table_percentage(df$HasCrCard))
## column
## 0 1
## 29.44706 70.55294
print(table_percentage(df$IsActiveMember))
## column
## 0 1
## 48.50515 51.49485
# Select specific columns
selected_columns <- df[, names(df) %in% c('Age', 'Balance', 'CreditScore', 'EstimatedSalary')]
# Reshape the data and create boxplots
selected_columns %>%
gather() %>%
ggplot(aes(value)) +
facet_wrap(~ key, scales = "free") +
geom_boxplot() +
theme_minimal()
## Warning: Removed 1 rows containing non-finite values (`stat_boxplot()`).
It appears customers age 40-50+ are exiting more than others.
Customers with a higher balance also seem more likely to exit, which could be tied to age and is not as statistically significant
Credit Score and Salary have little to no impact
There is a wide spread of tenure across exited customers, reflecting that age is also related.
# Select specific columns
cols <- df[, names(df) %in% c('Age', 'Balance', 'CreditScore', 'EstimatedSalary', 'Tenure', 'Exited')]
# Reshape the data and create boxplots
cols %>%
gather(key, value, -Exited) %>%
ggplot(aes(x = key, y = value, fill = factor(Exited))) +
geom_boxplot() +
scale_fill_manual(values = c("0" = "#97e4b2", "1" = "#FACBE1")) + # Customize fill colors
facet_wrap(~ key, scales = "free", ncol = 2) +
theme_minimal() +
labs(title = "Comparison of Variables Churned",
x = "Variable",
y = "Value")
## Warning: Removed 1 rows containing non-finite values (`stat_boxplot()`).
While the largest portion of customers are/were in France, Germany has the relatively highest volume of churn accounting for 39% of overall. This could likely be a geographic offering or policy change in that country that’s driven an exodus rather than an automatic predictor
Females have a lower share of total population, but a higher share of churned population
# Function to compare percentage splits for each category where Exited is 1 vs. 0
compare_exited_percentage <- function(data, column) {
data %>%
group_by(Exited, {{ column }}) %>%
summarise(count = n()) %>%
mutate(percentage = count / sum(count))
}
# Geography
compare_geography <- compare_exited_percentage(df, Geography)
## `summarise()` has grouped output by 'Exited'. You can override using the
## `.groups` argument.
print(compare_geography)
## # A tibble: 7 × 4
## # Groups: Exited [2]
## Exited Geography count percentage
## <dbl> <chr> <int> <dbl>
## 1 0 France 4204 0.528
## 2 0 Germany 1695 0.213
## 3 0 Spain 2064 0.259
## 4 0 <NA> 1 0.000126
## 5 1 France 810 0.397
## 6 1 Germany 815 0.400
## 7 1 Spain 413 0.203
# Gender
compare_gender <- compare_exited_percentage(df, Gender)
## `summarise()` has grouped output by 'Exited'. You can override using the
## `.groups` argument.
print(compare_gender)
## # A tibble: 4 × 4
## # Groups: Exited [2]
## Exited Gender count percentage
## <dbl> <chr> <int> <dbl>
## 1 0 Female 3405 0.428
## 2 0 Male 4559 0.572
## 3 1 Female 1139 0.559
## 4 1 Male 899 0.441
# Credit Card
compare_card <- compare_exited_percentage(df, HasCrCard)
## `summarise()` has grouped output by 'Exited'. You can override using the
## `.groups` argument.
print(compare_card)
## # A tibble: 5 × 4
## # Groups: Exited [2]
## Exited HasCrCard count percentage
## <dbl> <dbl> <int> <dbl>
## 1 0 0 2332 0.293
## 2 0 1 5631 0.707
## 3 0 NA 1 0.000126
## 4 1 0 613 0.301
## 5 1 1 1425 0.699
Difficult to see any trend on tenure here, so moving on to age later
# Create a histogram for Tenure for Exited = 1 and Exited = 0
ggplot(df, aes(x = Tenure, fill = factor(Exited))) +
geom_histogram(bins = 30, color = "white", alpha = 0.7, position = "identity") +
labs(title = "Tenure",
x = "Tenure",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_wrap(~ Exited, scales = "free") +
theme_minimal() +
scale_x_continuous(breaks = unique(df$Tenure))
Given the open values, this groups balance by $25k chunks. The data set has a max balance of just over 250k.
Output shows that customers with a balance of under 25K (likely 0K) or over 75k are most likely to churn, though for a model we would likely focus on the higher value customers assuming lower value balances are lower priority or insignificant. Because of this, we add a new feature of Balance75k, which will return a 1 if the balance is greater than 75k, and a 0 if less. We can add more later when trying out models.
# Define the breaks for binning (inclusive on the right)
breaks <- c(0, 25000, 50000, 75000, 100000, 125000, 150000, 175000, 200000, Inf)
# Create a new field 'Balance_Group' by binning the 'Balance' values
df <- df %>%
mutate(BalanceGroup = cut(Balance, breaks = breaks, labels = c("0k-25k", "25k-50k", "50k-75k", "75k-100k", "100k-125k", "125k-150k", "150k-175k", "175k-200k", "200k+"), include.lowest = TRUE, right = TRUE))
ggplot(df, aes(x = BalanceGroup, fill = factor(Exited))) +
geom_bar(position = "dodge", color = "white") +
labs(title = "Balance Group",
x = "Balance_Group",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_grid(Exited ~ ., scales = "free", switch = "y") +
theme_minimal()
df <- df %>%
mutate(Balance75k = ifelse(Balance >= 75000, 1, 0))
Assumes current or end-of-life products for relevance. Potentially shows that if a customer has 2 or more products they are less likely to churn
# Create a histogram for Product for Exited = 1 and Exited = 0
ggplot(df, aes(x = NumOfProducts, fill = factor(Exited))) +
geom_bar(bins = 30, color = "white", alpha = 0.7, position = "identity") +
labs(title = "Products",
x = "Number of Products",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_wrap(~ Exited, scales = "free") +
theme_minimal() +
scale_x_continuous(breaks = unique(df$NumOfProducts))
## Warning in geom_bar(bins = 30, color = "white", alpha = 0.7, position =
## "identity"): Ignoring unknown parameters: `bins`
Assumes current or end-of-life value for HasCrCard. This shows a pretty even split between churned vs not, and is likely not valuable to any models
# Create a histogram for Credit Card for Exited = 1 and Exited = 0
ggplot(df, aes(x = HasCrCard, fill = factor(Exited))) +
geom_bar(bins = 30, color = "white", alpha = 0.7, position = "identity") +
labs(title = "Has Credit Card",
x = "HasCrCard",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_wrap(~ Exited, scales = "free") +
theme_minimal() +
scale_x_continuous(breaks = unique(df$HasCrCard))
## Warning in geom_bar(bins = 30, color = "white", alpha = 0.7, position =
## "identity"): Ignoring unknown parameters: `bins`
## Warning: Removed 1 rows containing non-finite values (`stat_count()`).
Assumes current or end-of-life value for IsActiveMember. Shows that if they are an active member, they have more likelihood to stay as a customer - though I’d be hesitant to run with this incase products drop in the lead up to churn.
# Create a histogram for Credit Card for Exited = 1 and Exited = 0
ggplot(df, aes(x = IsActiveMember, fill = factor(Exited))) +
geom_bar(bins = 30, color = "white", alpha = 0.7, position = "identity") +
labs(title = "IsActiveMember",
x = "IsActiveMember",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_wrap(~ Exited, scales = "free") +
theme_minimal() +
scale_x_continuous(breaks = unique(df$IsActiveMember))
## Warning in geom_bar(bins = 30, color = "white", alpha = 0.7, position =
## "identity"): Ignoring unknown parameters: `bins`
## Warning: Removed 1 rows containing non-finite values (`stat_count()`).
Given the open values this groups balance by $25k chunks. The data set has a max balance of just under 200k.
There doesn’t appear to be any impact from Salary on churn.
# Define the breaks for binning (inclusive on the right)
breaks <- c(0, 25000, 50000, 75000, 100000, 125000, 150000, 175000, 200000, Inf)
# Create a new field 'Balance_Group' by binning the 'Balance' values
df <- df %>%
mutate(SalaryGroup = cut(EstimatedSalary, breaks = breaks, labels = c("0k-25k", "25k-50k", "50k-75k", "75k-100k", "100k-125k", "125k-150k", "150k-175k", "175k-200k", "200k+"), include.lowest = TRUE, right = TRUE))
ggplot(df, aes(x = SalaryGroup, fill = factor(Exited))) +
geom_bar(position = "dodge", color = "white") +
labs(title = "Salary Group",
x = "Salary Group",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_grid(Exited ~ ., scales = "free", switch = "y") +
theme_minimal() +
theme(axis.text.x = element_text(angle = -75, hjust = 1))
Age Groups 35-54 have the highest rate of churn, but for clarity, we’d want to look at this in smaller, possibly 5 year buckets
# Define the breaks for binning (inclusive on the right)
breaks <- c(18, 25, 35, 45, 55, 65, 75, 85, 95, Inf)
# Create a new field 'Age_Group' by binning the 'Age' values
df <- df %>%
mutate(AgeGroup = cut(Age, breaks = breaks, labels = c("18-24", "25-34", "35-44", "45-54", "55-64", "65-74", "75-84", "85-94", "95+"), include.lowest = TRUE, right = FALSE))
ggplot(df, aes(x = AgeGroup, fill = factor(Exited))) +
geom_bar(position = "dodge", color = "white") +
labs(title = "Age Group",
x = "Age Group",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_grid(Exited ~ ., scales = "free", switch = "y") +
theme_minimal() +
theme(axis.text.x = element_text(angle = -75, hjust = 1))
Churn expands at 35+, we can add a new binary that determines if the customers is 35+ (1) or not (0)
# Create breaks and labels
# Define age buckets
breaks <- c(18, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, Inf)
# Create a new column for age buckets
df <- df %>%
mutate(AgeGroup5s = cut(Age, breaks = breaks, labels = c("18-24", "25-29", "30-34", "35-39", "40-44",
"45-49", "50-54", "55-59", "60-64", "65-69",
"70-74", "75-79", "80-84", "85-89", "90-95", "95+"),
include.lowest = TRUE, right = FALSE))
ggplot(df, aes(x = AgeGroup5s, fill = factor(Exited))) +
geom_bar(position = "dodge", color = "white") +
labs(title = "Age Group",
x = "Age Group",
y = "Frequency") +
scale_fill_manual(values = c("#97e4b2", "#FACBE1")) +
facet_grid(Exited ~ ., scales = "free", switch = "y") +
theme_minimal() +
theme(axis.text.x = element_text(angle = -75, hjust = 1))
df <- df %>%
mutate(Age35 = ifelse(Age >= 35, 1, 0))
Overall:
# Generate a random identifier for each record
df <- df %>%
mutate(Id = stri_rand_strings(n = nrow(df), length = 10, pattern = "[A-Za-z0-9]"))
# Reorder columns to have Id as the first column, and drop irrelevant colums
df <- df %>%
select(Id, Exited, everything()) %>%
select(-CustomerId, -Surname, -EstimatedSalary, -SalaryGroup, -HasCrCard, -Age, -Balance, -BalanceGroup, -SalaryGroup, -AgeGroup, - AgeGroup5s)
#define function to scale values for CreditScore between 0 and 1
scale_values <- function(x){(x-min(x))/(max(x)-min(x))}
#scale values in CreditScore column to be between 0 and 1
df$CreditScore <- scale_values(df$CreditScore)
# Encode Geography and Gender
df$Geography <- as.numeric(factor(df$Geography, levels = c("Spain", "Germany", "France")))
df$Gender <- as.numeric(factor(df$Gender, levels = c("Male", "Female")))
# Print the updated dataframe
print(df)
## # A tibble: 10,002 × 10
## Id Exited Credi…¹ Geogr…² Gender Tenure NumOf…³ IsAct…⁴ Balan…⁵ Age35
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 lpgiMaDbNX 1 0.61 1 1 3 3 1 1 1
## 2 HnnPsBrQS4 1 0.466 3 2 3 1 1 1 1
## 3 6K6qstLSkB 1 0.626 1 2 5 1 1 1 1
## 4 QqI3WzePnR 0 1 3 2 3 1 1 1 1
## 5 3xYs17TXFq 0 0.382 3 1 9 1 1 1 1
## 6 cDE4mWEF8O 1 0.492 3 1 4 2 1 1 0
## 7 5tQJ9q0Vls 1 0.644 1 2 2 2 1 1 1
## 8 PUqkF0kwi5 1 0.712 1 2 1 1 1 1 0
## 9 FaLX91fKiZ 0 0.446 2 2 9 2 1 1 1
## 10 SPbprMkk7b 1 0.428 3 2 7 1 1 1 1
## # … with 9,992 more rows, and abbreviated variable names ¹CreditScore,
## # ²Geography, ³NumOfProducts, ⁴IsActiveMember, ⁵Balance75k