Sales and Marketing Analytics Project
Canon Tumwet
2025-08-24
#This script provides a comprehensive guide to performing data cleaning and exploratory data analysis (EDA) on a sales and marketing dataset # SECTION 1: Setup and Data Loading
1.1. Install and Load Necessary Packages
We will use tidyverse for data manipulation and
visualization, and skimr for quick data summaries.
library(tidyverse) # Includes dplyr, ggplot2, etc.── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
✔ dplyr 1.1.4 ✔ readr 2.1.5
✔ forcats 1.0.0 ✔ stringr 1.5.1
✔ ggplot2 3.5.2 ✔ tibble 3.2.1
✔ lubridate 1.9.4 ✔ tidyr 1.3.1
✔ purrr 1.0.4
── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()
ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errorslibrary(skimr) # For quick and comprehensive data summariesSet a seed for reproducibility of any random operations (though not many in this section)
set.seed(123)1.2. Load Sales and Marketing Dataset
df <- read_csv("sales_marketing_data.csv")Rows: 2500 Columns: 14
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr (4): gender, education, region, campaign_type
dbl (9): customer_id, age, income, marketing_spend, impressions, clicks, co...
date (1): date
ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.Display the first few rows to understand the data structure
print(head(df))# A tibble: 6 × 14
customer_id age gender income education region campaign_type marketing_spend
<dbl> <dbl> <chr> <dbl> <chr> <chr> <chr> <dbl>
1 1 56 Female 52778. High Sch… North Social Media 114.
2 2 69 Female 56382. Bachelors East Print Ad 72.5
3 3 46 Female 53334. High Sch… South Email 89.0
4 4 32 Female 69183. Bachelors West Email 101.
5 5 60 Male 35715. Bachelors East Social Media 89.5
6 6 25 Male 39845. Bachelors East Print Ad 127.
# ℹ 6 more variables: impressions <dbl>, clicks <dbl>, conversion_rate <dbl>,
# converted <dbl>, sales_amount <dbl>, date <date>Get a concise summary of the dataset, including data types and non-null counts
print(str(df))spc_tbl_ [2,500 × 14] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
$ customer_id : num [1:2500] 1 2 3 4 5 6 7 8 9 10 ...
$ age : num [1:2500] 56 69 46 32 60 25 38 56 36 40 ...
$ gender : chr [1:2500] "Female" "Female" "Female" "Female" ...
$ income : num [1:2500] 52778 56382 53334 69183 35715 ...
$ education : chr [1:2500] "High School" "Bachelors" "High School" "Bachelors" ...
$ region : chr [1:2500] "North" "East" "South" "West" ...
$ campaign_type : chr [1:2500] "Social Media" "Print Ad" "Email" "Email" ...
$ marketing_spend: num [1:2500] 114.3 72.5 89 101 89.5 ...
$ impressions : num [1:2500] 2886 1030 2709 4563 4368 ...
$ clicks : num [1:2500] 69 388 11 406 459 66 436 310 296 147 ...
$ conversion_rate: num [1:2500] 0.025 0.028 0.03 0.04 0.02 0.023 0.035 0.038 0.04 0.028 ...
$ converted : num [1:2500] 0 0 0 1 0 0 0 0 0 0 ...
$ sales_amount : num [1:2500] 0 0 0 242 0 ...
$ date : Date[1:2500], format: "2024-12-26" "2024-04-18" ...
- attr(*, "spec")=
.. cols(
.. customer_id = col_double(),
.. age = col_double(),
.. gender = col_character(),
.. income = col_double(),
.. education = col_character(),
.. region = col_character(),
.. campaign_type = col_character(),
.. marketing_spend = col_double(),
.. impressions = col_double(),
.. clicks = col_double(),
.. conversion_rate = col_double(),
.. converted = col_double(),
.. sales_amount = col_double(),
.. date = col_date(format = "")
.. )
- attr(*, "problems")=<externalptr>
NULLUse skimr for a more detailed and visually appealing summary
print(skim(df))── Data Summary ────────────────────────
Values
Name df
Number of rows 2500
Number of columns 14
_______________________
Column type frequency:
character 4
Date 1
numeric 9
________________________
Group variables None
── Variable type: character ────────────────────────────────────────────────────
skim_variable n_missing complete_rate min max empty n_unique whitespace
1 gender 0 1 4 6 0 3 0
2 education 0 1 3 11 0 4 0
3 region 0 1 4 5 0 4 0
4 campaign_type 0 1 5 12 0 4 0
── Variable type: Date ─────────────────────────────────────────────────────────
skim_variable n_missing complete_rate min max median
1 date 0 1 2024-01-01 2024-12-30 2024-06-25
n_unique
1 365
── Variable type: numeric ──────────────────────────────────────────────────────
skim_variable n_missing complete_rate mean sd p0
1 customer_id 0 1 1250. 722. 1
2 age 75 0.97 43.6 15.0 18
3 income 75 0.97 49807. 14989. 2349.
4 marketing_spend 0 1 101. 30.3 7.52
5 impressions 0 1 2774. 1306. 500
6 clicks 75 0.97 252. 143. 10
7 conversion_rate 0 1 0.0308 0.00720 0.02
8 converted 0 1 0.0308 0.173 0
9 sales_amount 0 1 7.47 44.1 0
p25 p50 p75 p100 hist
1 626. 1250. 1875. 2500 ▇▇▇▇▇
2 31 44 56 69 ▇▇▇▇▇
3 39812. 49958. 59931. 98646. ▁▃▇▃▁
4 80.3 101. 121. 203. ▁▅▇▃▁
5 1622. 2806 3886 4999 ▇▇▇▇▇
6 125 256 372 499 ▇▇▇▇▇
7 0.025 0.03 0.035 0.05 ▇▆▃▃▁
8 0 0 0 1 ▇▁▁▁▁
9 0 0 0 397. ▇▁▁▁▁#SECTION 2: Data Cleaning # 2.1. Handle Missing Values # From the skim() output, we can see missing values in ‘income’, ‘age’, and ‘clicks’. # For numerical columns, median imputation is often preferred over mean imputation because it is less sensitive to outliers.
print(colSums(is.na(df))) customer_id age gender income education
0 75 0 75 0
region campaign_type marketing_spend impressions clicks
0 0 0 0 75
conversion_rate converted sales_amount date
0 0 0 0 Impute ‘income’ with its median
Rationale: Income can be skewed, so median is a robust choice.
median_income <- median(df$income, na.rm = TRUE)
df$income[is.na(df$income)] <- median_incomeImpute ‘age’ with its median
Rationale: Age distribution might not be perfectly normal, median is safer.
median_age <- median(df$age, na.rm = TRUE)
df$age[is.na(df$age)] <- median_age
print(paste("Imputed missing 'age' with median:", median_age))[1] "Imputed missing 'age' with median: 44"Impute ‘clicks’ with its median
Rationale: Clicks data can also be skewed, median is a good choice.
median_clicks <- median(df$clicks, na.rm = TRUE)
df$clicks[is.na(df$clicks)] <- median_clicks
print(paste("Imputed missing 'clicks' with median:", median_clicks))[1] "Imputed missing 'clicks' with median: 256"print(colSums(is.na(df))) customer_id age gender income education
0 0 0 0 0
region campaign_type marketing_spend impressions clicks
0 0 0 0 0
conversion_rate converted sales_amount date
0 0 0 0 2.2. Correct Data Types
Ensure categorical variables are factors and date columns are in date format.
read_csv from tidyverse often does a good
job, but it’s good to explicitly check.
Convert character columns to factors if they represent categories
Rationale: Factors are R’s way of handling categorical variables, which is important for statistical modeling and some plotting functions.
df$gender <- as.factor(df$gender)
df$education <- as.factor(df$education)
df$region <- as.factor(df$region)
df$campaign_type <- as.factor(df$campaign_type)
df$converted <- as.factor(df$converted) # Target variable for classificationConvert ‘date’ column to Date type
Rationale: Proper date format allows for time-series analysis and filtering by date.
df$date <- as.Date(df$date)
print(str(df))spc_tbl_ [2,500 × 14] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
$ customer_id : num [1:2500] 1 2 3 4 5 6 7 8 9 10 ...
$ age : num [1:2500] 56 69 46 32 60 25 38 56 36 40 ...
$ gender : Factor w/ 3 levels "Female","Male",..: 1 1 1 1 2 2 1 2 1 1 ...
$ income : num [1:2500] 52778 56382 53334 69183 35715 ...
$ education : Factor w/ 4 levels "Bachelors","High School",..: 2 1 2 1 1 1 1 3 3 1 ...
$ region : Factor w/ 4 levels "East","North",..: 2 1 3 4 1 1 1 2 3 1 ...
$ campaign_type : Factor w/ 4 levels "Email","Print Ad",..: 3 2 1 1 3 2 4 2 1 2 ...
$ marketing_spend: num [1:2500] 114.3 72.5 89 101 89.5 ...
$ impressions : num [1:2500] 2886 1030 2709 4563 4368 ...
$ clicks : num [1:2500] 69 388 11 406 459 66 436 310 296 147 ...
$ conversion_rate: num [1:2500] 0.025 0.028 0.03 0.04 0.02 0.023 0.035 0.038 0.04 0.028 ...
$ converted : Factor w/ 2 levels "0","1": 1 1 1 2 1 1 1 1 1 1 ...
$ sales_amount : num [1:2500] 0 0 0 242 0 ...
$ date : Date[1:2500], format: "2024-12-26" "2024-04-18" ...
- attr(*, "spec")=
.. cols(
.. customer_id = col_double(),
.. age = col_double(),
.. gender = col_character(),
.. income = col_double(),
.. education = col_character(),
.. region = col_character(),
.. campaign_type = col_character(),
.. marketing_spend = col_double(),
.. impressions = col_double(),
.. clicks = col_double(),
.. conversion_rate = col_double(),
.. converted = col_double(),
.. sales_amount = col_double(),
.. date = col_date(format = "")
.. )
- attr(*, "problems")=<externalptr>
NULLSECTION 3: Exploratory Data Analysis (EDA)
3.1. Descriptive Statistics for Numerical Variables
Get summary statistics for key numerical columns.
Rationale: Provides a quick overview of central tendency, dispersion, and range.
library(psych)
Attaching package: 'psych'The following objects are masked from 'package:ggplot2':
%+%, alphaselected_variables<-df %>%
select(income,marketing_spend,impressions,clicks,sales_amount)
describe(selected_variables) vars n mean sd median trimmed mad min
income 1 2500 49811.57 14762.85 49958.21 49860.72 14392.47 2349.44
marketing_spend 2 2500 100.73 30.30 100.61 100.63 30.17 7.52
impressions 3 2500 2773.97 1305.95 2806.00 2783.16 1668.67 500.00
clicks 4 2500 252.15 140.55 256.00 251.93 177.91 10.00
sales_amount 5 2500 7.47 44.11 0.00 0.00 0.00 0.00
max range skew kurtosis se
income 98646.39 96296.95 -0.02 0.10 295.26
marketing_spend 202.87 195.35 0.04 0.04 0.61
impressions 4999.00 4499.00 -0.06 -1.21 26.12
clicks 499.00 489.00 -0.01 -1.16 2.81
sales_amount 397.36 397.36 6.21 38.91 0.883.2. Distribution of Key Numerical Variables
Histograms and density plots help visualize the shape of distributions.
Rationale: Identify skewness, outliers, and overall data spread.
Distribution of Income
plot_income_dist <- ggplot(df, aes(x = income)) +
geom_histogram(binwidth = 5000, fill = "skyblue", color = "black") +
labs(title = "Distribution of Customer Income", x = "Income", y = "Frequency")
print(plot_income_dist)
ggsave("income_distribution.png", plot_income_dist, width = 8, height = 6)
print("Saved: income_distribution.png")[1] "Saved: income_distribution.png"Distribution of Sales Amount (for converted customers)
plot_sales_dist <- ggplot(filter(df, converted == 1), aes(x = sales_amount)) +
geom_histogram(binwidth = 25, fill = "lightgreen", color = "black") +
labs(title = "Distribution of Sales Amount (Converted Customers)", x = "Sales Amount", y = "Frequency")
print(plot_sales_dist)
ggsave("sales_amount_distribution.png", plot_sales_dist, width = 8, height = 6)
print("Saved: sales_amount_distribution.png")[1] "Saved: sales_amount_distribution.png"3.3. Relationship Between Numerical Variables
Scatter plots and correlation matrices help understand relationships.
Rationale: Identify potential predictors and multicollinearity.
Scatter plot: Marketing Spend vs. Clicks
plot_spend_clicks <- ggplot(df, aes(x = marketing_spend, y = clicks)) +
geom_point(alpha = 0.6, color = "darkblue") +
labs(title = "Marketing Spend vs. Clicks", x = "Marketing Spend", y = "Clicks")
print(plot_spend_clicks)
ggsave("marketing_spend_vs_clicks.png", plot_spend_clicks, width = 8, height = 6)
print("Saved: marketing_spend_vs_clicks.png")[1] "Saved: marketing_spend_vs_clicks.png"Correlation Matrix (for selected numerical variables)
Rationale: Quantify linear relationships between variables.
Note: cor() requires a matrix of numerical values.
num_vars <- df %>% select(income, marketing_spend, impressions, clicks, sales_amount)
print(cor(num_vars)) income marketing_spend impressions clicks
income 1.000000000 -0.007837892 0.020403122 0.010405013
marketing_spend -0.007837892 1.000000000 -0.005066432 -0.008536314
impressions 0.020403122 -0.005066432 1.000000000 -0.048113517
clicks 0.010405013 -0.008536314 -0.048113517 1.000000000
sales_amount 0.025573430 -0.014129812 0.014787840 -0.005072715
sales_amount
income 0.025573430
marketing_spend -0.014129812
impressions 0.014787840
clicks -0.005072715
sales_amount 1.0000000003.4. Analysis of Categorical Variables
Bar plots are excellent for visualizing the counts or proportions of categories.
Rationale: Understand the composition of customer segments or campaign types.
Count of Customers by Gender
plot_gender_counts <- ggplot(df, aes(x = gender, fill = gender)) +
geom_bar() +
labs(title = "Customer Count by Gender", x = "Gender", y = "Count")
print(plot_gender_counts)
ggsave("gender_counts.png", plot_gender_counts, width = 8, height = 6)
print("Saved: gender_counts.png")[1] "Saved: gender_counts.png"Sales Amount by Campaign Type (using box plots for distribution)
Rationale: Compare sales performance across different marketing channels.
plot_sales_by_campaign <- ggplot(df, aes(x = campaign_type, y = sales_amount, fill = campaign_type)) +
geom_boxplot() +
labs(title = "Sales Amount by Campaign Type", x = "Campaign Type", y = "Sales Amount")
print(plot_sales_by_campaign)
Conversion Rate by Education Level
Rationale: See if education influences conversion.
conversion_by_education <- df %>%
group_by(education) %>%
summarise(conversion_rate = mean(converted == 1)) # Calculate mean of TRUE (1) for conversion rate
plot_conversion_by_education <- ggplot(conversion_by_education, aes(x = education, y = conversion_rate, fill = education)) +
geom_bar(stat = "identity") +
labs(title = "Conversion Rate by Education Level", x = "Education Level", y = "Conversion Rate")
print(plot_conversion_by_education)
ggsave("conversion_by_education.png", plot_conversion_by_education, width = 8, height = 6)
print("Saved: conversion_by_education.png")[1] "Saved: conversion_by_education.png"3.5. Time-Series Analysis (Basic)
Analyze trends over time, e.g., daily sales.
Rationale: Identify seasonality, growth, or decline patterns.
daily_sales <- df %>%
group_by(date) %>%
summarise(total_sales = sum(sales_amount))
plot_daily_sales <- ggplot(daily_sales, aes(x = date, y = total_sales)) +
geom_line(color = "purple") +
labs(title = "Daily Total Sales Amount", x = "Date", y = "Total Sales")
print(plot_daily_sales)
ggsave("daily_total_sales.png", plot_daily_sales, width = 10, height = 6)
print("Saved: daily_total_sales.png")[1] "Saved: daily_total_sales.png"SECTION 4: Predictive Modeling
This section focuses on building machine learning models to predict key sales and marketing outcomes.
We will demonstrate both regression (predicting continuous sales amount) and classification
(predicting whether a customer converts).
4.1. Install and Load Necessary Packages for Modeling
We will use caret for streamlined model training and
evaluation, and randomForest for a popular ensemble
model.
install.packages(“caret”)
install.packages(“randomForest”)
install.packages(“e1071”) # Often a dependency for caret
install.packages(“glmnet”) # For Lasso/Ridge Regression
library(caret) # For model training and evaluation workflowLoading required package: lattice
Attaching package: 'caret'The following object is masked from 'package:purrr':
liftlibrary(randomForest) # For Random Forest modelrandomForest 4.7-1.2Type rfNews() to see new features/changes/bug fixes.
Attaching package: 'randomForest'The following object is masked from 'package:psych':
outlierThe following object is masked from 'package:dplyr':
combineThe following object is masked from 'package:ggplot2':
marginlibrary(e1071) # Contains SVM and other functions, often a caret dependency
library(glmnet) # For regularized regression (Lasso, Ridge)Loading required package: Matrix
Attaching package: 'Matrix'The following objects are masked from 'package:tidyr':
expand, pack, unpackLoaded glmnet 4.1-84.2. Prepare Data for Modeling
We need to split the data into training and testing sets.
For regression, the target is sales_amount.
For classification, the target is converted.
Remove customer_id and date as they are
not features for modeling.
conversion_rate is also a calculated outcome, not a
direct feature for prediction.
model_df <- df %>% select(-customer_id, -date, -conversion_rate)Ensure all categorical variables are factors for
caret
(Already done in Data Cleaning, but good to re-confirm)
Define target variables and features
For regression: predict sales_amount
For classification: predict converted
— 4.2.1. Data Split for Regression (Predicting Sales Amount) —
We only consider rows where sales_amount > 0 for regression, as 0 sales means no conversion.
Rationale: Predicting 0 sales for non-converted customers is a classification problem.
Here, we focus on predicting the value of sales for converted customers.
regression_df <- model_df %>% filter(sales_amount > 0)Create training and testing sets for regression
createDataPartition from caret ensures
stratified sampling for classification, but works for regression
too.
set.seed(123) # For reproducibility
regression_index <- createDataPartition(regression_df$sales_amount, p = 0.8, list = FALSE)
regression_train <- regression_df[regression_index, ]
regression_test <- regression_df[-regression_index, ]
print(paste("Regression Training set size:", nrow(regression_train)))[1] "Regression Training set size: 64"print(paste("Regression Test set size:", nrow(regression_test)))[1] "Regression Test set size: 13"4.2.2. Data Split for Classification (Predicting Conversion)
Here, the target is converted (0 or 1).
set.seed(123) # For reproducibility
classification_index <- createDataPartition(model_df$converted, p = 0.8, list = FALSE)
classification_train <- model_df[classification_index, ]
classification_test <- model_df[-classification_index, ]
print(paste("Classification Training set size:", nrow(classification_train)))[1] "Classification Training set size: 2001"print(paste("Classification Test set size:", nrow(classification_test)))[1] "Classification Test set size: 499"4.3. Supervised Learning - Regression Models (Predicting Sales Amount)
We will use caret for training and evaluating
models.
trainControl sets up the resampling method (e.g.,
cross-validation).
method = "cv" for cross-validation,
number = 10 for 10-fold CV.
verboseIter = FALSE to suppress detailed output during
training.
train_control <- trainControl(method = "cv", number = 10, verboseIter = FALSE)Model 1: Linear Regression
Rationale: A simple, interpretable baseline model. Assumes a linear relationship.
set.seed(123)
linear_model <- train(sales_amount ~ ., data = regression_train, method = "lm", trControl = train_control)
print(linear_model)Linear Regression
64 samples
10 predictors
No pre-processing
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 58, 58, 57, 57, 57, 60, ...
Resampling results:
RMSE Rsquared MAE
93.14125 0.1716143 80.27489
Tuning parameter 'intercept' was held constant at a value of TRUEMake predictions and evaluate
linear_predictions <- predict(linear_model, newdata = regression_test)
linear_metrics <- postResample(linear_predictions, regression_test$sales_amount)
print("Linear Regression Metrics:")[1] "Linear Regression Metrics:"print(linear_metrics) RMSE Rsquared MAE
68.54136400 0.05299308 53.05824823 Model 2: Random Forest Regression
Rationale: Powerful ensemble method, robust to outliers, captures non-linear relationships,and provides good accuracy. Less prone to overfitting than single decision trees.
set.seed(123)
rf_model <- train(sales_amount ~ ., data = regression_train, method = "rf", trControl = train_control)
print(rf_model)Random Forest
64 samples
10 predictors
No pre-processing
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 58, 58, 57, 57, 57, 60, ...
Resampling results across tuning parameters:
mtry RMSE Rsquared MAE
2 82.13044 0.1219000 70.31225
9 84.51320 0.1079994 71.91191
17 85.16276 0.1174431 71.79690
RMSE was used to select the optimal model using the smallest value.
The final value used for the model was mtry = 2.Make predictions and evaluate
rf_predictions <- predict(rf_model, newdata = regression_test)
rf_metrics <- postResample(rf_predictions, regression_test$sales_amount)
print("Random Forest Regression Metrics:")[1] "Random Forest Regression Metrics:"print(rf_metrics) RMSE Rsquared MAE
66.76279754 0.07003723 54.26066152 4.4. Supervised Learning - Classification Models (Predicting Conversion)
Model 1: Logistic Regression
Rationale: A fundamental classification algorithm for binary outcomes. Interpretable.
family = "binomial" specifies logistic regression.
set.seed(123)
logistic_model <- train(converted ~ ., data = classification_train, method = "glm", family = "binomial", trControl = train_control)
print(logistic_model)Generalized Linear Model
2001 samples
10 predictor
2 classes: '0', '1'
No pre-processing
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 1801, 1801, 1800, 1801, 1801, 1801, ...
Resampling results:
Accuracy Kappa
1 1 Make predictions and evaluate
logistic_predictions <- predict(logistic_model, newdata = classification_test)
logistic_metrics <- confusionMatrix(logistic_predictions, classification_test$converted)
print("Logistic Regression Confusion Matrix:")[1] "Logistic Regression Confusion Matrix:"print(logistic_metrics)Confusion Matrix and Statistics
Reference
Prediction 0 1
0 484 0
1 0 15
Accuracy : 1
95% CI : (0.9926, 1)
No Information Rate : 0.9699
P-Value [Acc > NIR] : 2.43e-07
Kappa : 1
Mcnemar's Test P-Value : NA
Sensitivity : 1.0000
Specificity : 1.0000
Pos Pred Value : 1.0000
Neg Pred Value : 1.0000
Prevalence : 0.9699
Detection Rate : 0.9699
Detection Prevalence : 0.9699
Balanced Accuracy : 1.0000
'Positive' Class : 0
— Model 2: Support Vector Machine (SVM) —
Rationale: Effective in high-dimensional spaces and cases where number of dimensions
is greater than the number of samples. Can use different kernels to capture non-linearities.
set.seed(123)
svm_model <- train(converted ~ ., data = classification_train, method = "svmRadial", trControl = train_control)
print(svm_model)Support Vector Machines with Radial Basis Function Kernel
2001 samples
10 predictor
2 classes: '0', '1'
No pre-processing
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 1801, 1801, 1800, 1801, 1801, 1801, ...
Resampling results across tuning parameters:
C Accuracy Kappa
0.25 0.9989975 0.9813071
0.50 1.0000000 1.0000000
1.00 1.0000000 1.0000000
Tuning parameter 'sigma' was held constant at a value of 0.0387194
Accuracy was used to select the optimal model using the largest value.
The final values used for the model were sigma = 0.0387194 and C = 0.5.Make predictions and evaluate
svm_predictions <- predict(svm_model, newdata = classification_test)
svm_metrics <- confusionMatrix(svm_predictions, classification_test$converted)
print("SVM Confusion Matrix:")[1] "SVM Confusion Matrix:"print(svm_metrics)Confusion Matrix and Statistics
Reference
Prediction 0 1
0 484 0
1 0 15
Accuracy : 1
95% CI : (0.9926, 1)
No Information Rate : 0.9699
P-Value [Acc > NIR] : 2.43e-07
Kappa : 1
Mcnemar's Test P-Value : NA
Sensitivity : 1.0000
Specificity : 1.0000
Pos Pred Value : 1.0000
Neg Pred Value : 1.0000
Prevalence : 0.9699
Detection Rate : 0.9699
Detection Prevalence : 0.9699
Balanced Accuracy : 1.0000
'Positive' Class : 0
— Model 3: K-Nearest Neighbors (KNN) —
Rationale: Non-parametric, instance-based learning. Simple to understand, but can be
computationally expensive for large datasets. Useful for understanding local patterns.
set.seed(123)
knn_model <- train(converted ~ ., data = classification_train, method = "knn", trControl = train_control)
print(knn_model)k-Nearest Neighbors
2001 samples
10 predictor
2 classes: '0', '1'
No pre-processing
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 1801, 1801, 1800, 1801, 1801, 1801, ...
Resampling results across tuning parameters:
k Accuracy Kappa
5 0.9690198 0
7 0.9690198 0
9 0.9690198 0
Accuracy was used to select the optimal model using the largest value.
The final value used for the model was k = 9.Make predictions and evaluate
knn_predictions <- predict(knn_model, newdata = classification_test)
knn_metrics <- confusionMatrix(knn_predictions, classification_test$converted)
print("KNN Confusion Matrix:")[1] "KNN Confusion Matrix:"print(knn_metrics)Confusion Matrix and Statistics
Reference
Prediction 0 1
0 484 15
1 0 0
Accuracy : 0.9699
95% CI : (0.9509, 0.9831)
No Information Rate : 0.9699
P-Value [Acc > NIR] : 0.5680975
Kappa : 0
Mcnemar's Test P-Value : 0.0003006
Sensitivity : 1.0000
Specificity : 0.0000
Pos Pred Value : 0.9699
Neg Pred Value : NaN
Prevalence : 0.9699
Detection Rate : 0.9699
Detection Prevalence : 1.0000
Balanced Accuracy : 0.5000
'Positive' Class : 0
SECTION 5: Unsupervised Learning (Clustering)
Here, we will use K-Means clustering to segment customers based on their demographics and behavior.
5.1. Prepare Data for Clustering
For clustering, we typically use numerical features and scale them to ensure all features
contribute equally to the distance calculations.
Select numerical features for clustering (excluding target variables and IDs)
clustering_df <- df %>%
select(age, income, marketing_spend, impressions, clicks, sales_amount) %>%
na.omit() # K-Means cannot handle missing valuesScale the data
Rationale: Features with larger scales would dominate the distance calculations without scaling.
scaled_clustering_df <- scale(clustering_df)
print("\n--- SECTION 5: Unsupervised Learning (Clustering) ---")[1] "\n--- SECTION 5: Unsupervised Learning (Clustering) ---"print("Data prepared and scaled for clustering.")[1] "Data prepared and scaled for clustering."5.2. K-Means Clustering
Determine optimal number of clusters (K) using the Elbow Method.
Rationale: The Elbow Method helps find a good balance between minimizing within-cluster sum of squares
and keeping the number of clusters manageable.
wss <- (nrow(scaled_clustering_df)-1)*sum(apply(scaled_clustering_df,2,var))
for (i in 2:10) wss[i] <- sum(kmeans(scaled_clustering_df, centers=i)$withinss)
plot(1:10, wss, type="b", xlab="Number of Clusters",
ylab="Within groups sum of squares",
main="Elbow Method for K-Means Clustering")
Choose K (e.g., K=3 based on a hypothetical elbow point)
Rationale: Let’s assume 3 distinct customer segments for demonstration.
k_clusters <- 3
set.seed(123)
kmeans_result <- kmeans(scaled_clustering_df, centers = k_clusters, nstart = 25)Add cluster assignments back to the original (non-scaled) dataframe for interpretation
df_clustered <- df %>%
filter(!is.na(income) & !is.na(age) & !is.na(clicks)) # Filter to match clustering_df
df_clustered$cluster <- as.factor(kmeans_result$cluster)
print(paste("K-Means Clustering performed with K =", k_clusters))[1] "K-Means Clustering performed with K = 3"print("Cluster sizes:")[1] "Cluster sizes:"print(table(df_clustered$cluster))
1 2 3
1261 69 1170 5.3. Interpret Clusters
Analyze the characteristics of each cluster by looking at the mean values of features.
Rationale: This helps in understanding what defines each customer segment.
print("\nMean values of features per cluster:")[1] "\nMean values of features per cluster:"print(df_clustered %>%
group_by(cluster) %>%
summarise(across(c(age, income, marketing_spend, impressions, clicks, sales_amount), mean)))# A tibble: 3 × 7
cluster age income marketing_spend impressions clicks sales_amount
<fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 1 44.0 50193. 99.4 2519. 366. 0.278
2 2 42.1 51305. 99.0 2948. 249. 257.
3 3 43.3 49312. 102. 3039. 130. 0.502You can also analyze categorical features per cluster using
table() or prop.table()
print("\nGender distribution per cluster:")[1] "\nGender distribution per cluster:"print(prop.table(table(df_clustered$cluster, df_clustered$gender), margin = 1))
Female Male Other
1 0.47581285 0.48374306 0.04044409
2 0.47826087 0.52173913 0.00000000
3 0.47521368 0.48034188 0.04444444print("\nCampaign Type distribution per cluster:")[1] "\nCampaign Type distribution per cluster:"print(prop.table(table(df_clustered$cluster, df_clustered$campaign_type), margin = 1))
Email Print Ad Social Media TV Ad
1 0.4147502 0.1562252 0.2862807 0.1427439
2 0.4202899 0.1739130 0.2898551 0.1159420
3 0.4000000 0.1572650 0.2965812 0.1461538print("\nSales and Marketing R Script (Predictive Modeling and Unsupervised Learning) completed.")[1] "\nSales and Marketing R Script (Predictive Modeling and Unsupervised Learning) completed."