Introduction:

The goal of this report is to identify which online shopping behaviors are most strongly associated with a customer’s decision to make a purchase. This data came from an e-commerce website, to determine which factors affect the probability of a purchase (revenue). The main research questions are which behavioral variables best predict whether or not a user will make a purchase, and how can these insights be used to help web design in order to bring more sales in.

Materials:

The dataset contains 12,330 sessions and 18 variables. Each row represents a unique browsing session. The response variable, revenue, indicates whether or not a purchase occurred. Some key predictor variables are page values, exit rates, bounce rates, product related duration, visitor type, month of visit, and weekend visits.

Methodology and analysis:

We applied a multiple logistic regression model, with the dependent variable being revenue.

Results and conclusions:

The final model retained significant predictors being bounce rates, exit rates, page values, visitor type, month of visit, product related duration, and weekend visit, with all p values < .05. The final AIC was 7203, very similar to that of the original model. The AUC was .8969, meaning the model did a great job in classification accuracy. The Hosmer-Lemeshow test rejected a perfect fit, but with a 12,000+ observation data set this is to be expected. Overall, page values and product related duration reflect the depth of engagement and are strong indicators of conversion. High exit rates highly reduced likelihood of a purchase, signaling how teams can improve their site. Month of visit indicated November as the highest month with purchase conversions. Returning visitors may research prior to purchasing, indicating a need for re targeting for customers. The final model explains consumers purchase behavior with an AUC of near 0.9, indicating that engagement based metrics like page value and product time are the most critical drivers of purchase conversions.

General Discussion:

This analysis provides insight into e-commerce behavior using web optimization, marketing timing and user retention. Some improvements are that the data set lacks demographic and pricing information, limiting some of the explanatory power within the model. Some variables are also skewed, which may distort effect elements. Incorporating interaction terms, cross-validation techniques, or supervised techniques can also be used to yield better results.

set.seed(321)
knitr::opts_chunk$set(message = FALSE, warning = FALSE)


library(tidyverse)         
## ── 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   3.5.1     ✔ tibble    3.2.1
## ✔ lubridate 1.9.3     ✔ tidyr     1.3.1
## ✔ purrr     1.0.2     
## ── 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 errors
library(broom)             
library(MASS)              
## 
## Attaching package: 'MASS'
## 
## The following object is masked from 'package:dplyr':
## 
##     select
library(car)               
## Warning: package 'car' was built under R version 4.4.2
## Loading required package: carData
## Warning: package 'carData' was built under R version 4.4.2
## 
## Attaching package: 'car'
## 
## The following object is masked from 'package:dplyr':
## 
##     recode
## 
## The following object is masked from 'package:purrr':
## 
##     some
library(ResourceSelection) 
## Warning: package 'ResourceSelection' was built under R version 4.4.3
## ResourceSelection 0.3-6   2023-06-27
library(pROC)              
## Type 'citation("pROC")' for a citation.
## 
## Attaching package: 'pROC'
## 
## The following objects are masked from 'package:stats':
## 
##     cov, smooth, var
library(pander)             
## Warning: package 'pander' was built under R version 4.4.3
data_path <- "C:/Users/rg03/Downloads/sta321/online_shoppers_intention.csv"



dat <- read.csv(data_path, stringsAsFactors = FALSE)



dat$Revenue <- factor(dat$Revenue, levels = c(FALSE, TRUE), labels = c("No", "Yes"))


glimpse(dat)
## Rows: 12,330
## Columns: 18
## $ Administrative          <int> 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 2…
## $ Administrative_Duration <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5…
## $ Informational           <int> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ Informational_Duration  <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ ProductRelated          <int> 1, 2, 1, 2, 10, 19, 1, 0, 2, 3, 3, 16, 7, 6, 2…
## $ ProductRelated_Duration <dbl> 0.000000, 64.000000, 0.000000, 2.666667, 627.5…
## $ BounceRates             <dbl> 0.200000000, 0.000000000, 0.200000000, 0.05000…
## $ ExitRates               <dbl> 0.200000000, 0.100000000, 0.200000000, 0.14000…
## $ PageValues              <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ SpecialDay              <dbl> 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.4, 0.0, 0.8, 0…
## $ Month                   <chr> "Feb", "Feb", "Feb", "Feb", "Feb", "Feb", "Feb…
## $ OperatingSystems        <int> 1, 2, 4, 3, 3, 2, 2, 1, 2, 2, 1, 1, 1, 2, 3, 1…
## $ Browser                 <int> 1, 2, 1, 2, 3, 2, 4, 2, 2, 4, 1, 1, 1, 5, 2, 1…
## $ Region                  <int> 1, 1, 9, 2, 1, 1, 3, 1, 2, 1, 3, 4, 1, 1, 3, 9…
## $ TrafficType             <int> 1, 2, 3, 4, 4, 3, 3, 5, 3, 2, 3, 3, 3, 3, 3, 3…
## $ VisitorType             <chr> "Returning_Visitor", "Returning_Visitor", "Ret…
## $ Weekend                 <lgl> FALSE, FALSE, FALSE, FALSE, TRUE, FALSE, FALSE…
## $ Revenue                 <fct> No, No, No, No, No, No, No, No, No, No, No, No…
summary(dat$Revenue)
##    No   Yes 
## 10422  1908
use_vars <- c(
"Administrative_Duration", "Informational_Duration", "ProductRelated_Duration",
"BounceRates", "ExitRates", "PageValues",           
"Month", "VisitorType", "Weekend"                   
)



dat2 <- dat %>%
dplyr::select(Revenue, dplyr::all_of(use_vars)) %>%
na.omit()



dat2 <- dat2 %>%
mutate(
Month = factor(Month),
VisitorType = factor(VisitorType),
Weekend = factor(Weekend)  
)



summary(dat2)
##  Revenue     Administrative_Duration Informational_Duration
##  No :10422   Min.   :   0.00         Min.   :   0.00       
##  Yes: 1908   1st Qu.:   0.00         1st Qu.:   0.00       
##              Median :   7.50         Median :   0.00       
##              Mean   :  80.82         Mean   :  34.47       
##              3rd Qu.:  93.26         3rd Qu.:   0.00       
##              Max.   :3398.75         Max.   :2549.38       
##                                                            
##  ProductRelated_Duration  BounceRates         ExitRates         PageValues     
##  Min.   :    0.0         Min.   :0.000000   Min.   :0.00000   Min.   :  0.000  
##  1st Qu.:  184.1         1st Qu.:0.000000   1st Qu.:0.01429   1st Qu.:  0.000  
##  Median :  598.9         Median :0.003112   Median :0.02516   Median :  0.000  
##  Mean   : 1194.8         Mean   :0.022191   Mean   :0.04307   Mean   :  5.889  
##  3rd Qu.: 1464.2         3rd Qu.:0.016813   3rd Qu.:0.05000   3rd Qu.:  0.000  
##  Max.   :63973.5         Max.   :0.200000   Max.   :0.20000   Max.   :361.764  
##                                                                                
##      Month                 VisitorType     Weekend    
##  May    :3364   New_Visitor      : 1694   FALSE:9462  
##  Nov    :2998   Other            :   85   TRUE :2868  
##  Mar    :1907   Returning_Visitor:10551               
##  Dec    :1727                                         
##  Oct    : 549                                         
##  Sep    : 448                                         
##  (Other):1337
pairs(
~ Administrative_Duration + Informational_Duration +
ProductRelated_Duration + BounceRates + ExitRates + PageValues,
data = dat2,
main = "Scatterplot Matrix of Continuous Predictors"
)

num_df <- dat2 %>% dplyr::select(where(is.numeric))
if (ncol(num_df) > 1) {
round(cor(num_df), 2)
}
##                         Administrative_Duration Informational_Duration
## Administrative_Duration                    1.00                   0.24
## Informational_Duration                     0.24                   1.00
## ProductRelated_Duration                    0.36                   0.35
## BounceRates                               -0.14                  -0.07
## ExitRates                                 -0.21                  -0.11
## PageValues                                 0.07                   0.03
##                         ProductRelated_Duration BounceRates ExitRates
## Administrative_Duration                    0.36       -0.14     -0.21
## Informational_Duration                     0.35       -0.07     -0.11
## ProductRelated_Duration                    1.00       -0.18     -0.25
## BounceRates                               -0.18        1.00      0.91
## ExitRates                                 -0.25        0.91      1.00
## PageValues                                 0.05       -0.12     -0.17
##                         PageValues
## Administrative_Duration       0.07
## Informational_Duration        0.03
## ProductRelated_Duration       0.05
## BounceRates                  -0.12
## ExitRates                    -0.17
## PageValues                    1.00
full.model <- glm(
Revenue ~ Administrative_Duration + Informational_Duration +
ProductRelated_Duration + BounceRates + ExitRates +
PageValues + Month + VisitorType + Weekend,
data = dat2, family = binomial(link = "logit")
)

pander(summary(full.model)$coef,
caption = "Full Model — Inferential Statistics")
Full Model — Inferential Statistics (continued below)
  Estimate Std. Error z value
(Intercept) -1.656 0.1739 -9.527
Administrative_Duration -7.046e-05 0.0001634 -0.4311
Informational_Duration 0.000219 0.0001822 1.202
ProductRelated_Duration 0.0001013 1.465e-05 6.914
BounceRates -3.808 3.252 -1.171
ExitRates -16.2 2.347 -6.905
PageValues 0.08223 0.002402 34.23
MonthDec -0.6181 0.1818 -3.401
MonthFeb -1.81 0.6354 -2.848
MonthJul 0.07887 0.2183 0.3613
MonthJune -0.3071 0.2751 -1.116
MonthMar -0.5278 0.1794 -2.942
MonthMay -0.5877 0.1692 -3.474
MonthNov 0.5368 0.1622 3.309
MonthOct -0.01127 0.2016 -0.05589
MonthSep -0.01799 0.2123 -0.0847
VisitorTypeOther -0.5364 0.5245 -1.023
VisitorTypeReturning_Visitor -0.3129 0.08512 -3.676
WeekendTRUE 0.09889 0.07086 1.396
  Pr(>|z|)
(Intercept) 1.614e-21
Administrative_Duration 0.6664
Informational_Duration 0.2293
ProductRelated_Duration 4.726e-12
BounceRates 0.2416
ExitRates 5.011e-12
PageValues 8.675e-257
MonthDec 0.0006725
MonthFeb 0.004398
MonthJul 0.7179
MonthJune 0.2643
MonthMar 0.003256
MonthMay 0.0005134
MonthNov 0.000935
MonthOct 0.9554
MonthSep 0.9325
VisitorTypeOther 0.3064
VisitorTypeReturning_Visitor 0.0002369
WeekendTRUE 0.1628 
vif(full.model)
##                             GVIF Df GVIF^(1/(2*Df))
## Administrative_Duration 1.158832  1        1.076491
## Informational_Duration  1.144643  1        1.069880
## ProductRelated_Duration 1.328126  1        1.152444
## BounceRates             1.984528  1        1.408733
## ExitRates               2.102475  1        1.449991
## PageValues              1.061579  1        1.030330
## Month                   1.107538  9        1.005691
## VisitorType             1.128430  2        1.030668
## Weekend                 1.011496  1        1.005732
reduced.model <- glm(
Revenue ~ BounceRates + ExitRates + PageValues + VisitorType,
data = dat2, family = binomial(link = "logit")
)

pander(summary(reduced.model)$coef,
caption = "Reduced Model — Inferential Statistics")
Reduced Model — Inferential Statistics
  Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.593 0.08097 -19.67 3.976e-86
BounceRates -0.1066 3.106 -0.03431 0.9726
ExitRates -21.26 2.256 -9.422 4.415e-21
PageValues 0.07907 0.002317 34.13 2.792e-255
VisitorTypeOther -0.6119 0.5099 -1.2 0.2301
VisitorTypeReturning_Visitor -0.1581 0.08121 -1.947 0.05154 
final.model <- stepAIC(
reduced.model,
scope = list(lower = formula(reduced.model), upper = formula(full.model)),
direction = "forward",
trace = 0
)

pander(summary(final.model)$coef,
caption = "Final Model (Stepwise Forward) — Inferential Statistics")
Final Model (Stepwise Forward) — Inferential Statistics
  Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.666 0.1729 -9.636 5.66e-22
BounceRates -3.788 3.245 -1.168 0.243
ExitRates -16.17 2.336 -6.921 4.5e-12
PageValues 0.08227 0.002399 34.29 1.14e-257
VisitorTypeOther -0.5388 0.5252 -1.026 0.3049
VisitorTypeReturning_Visitor -0.3099 0.08507 -3.643 0.0002692
MonthDec -0.6126 0.1817 -3.372 0.0007462
MonthFeb -1.812 0.6357 -2.85 0.004373
MonthJul 0.08542 0.2182 0.3915 0.6954
MonthJune -0.309 0.2752 -1.123 0.2614
MonthMar -0.5232 0.1793 -2.917 0.003533
MonthMay -0.5849 0.1692 -3.458 0.0005451
MonthNov 0.5389 0.1621 3.324 0.0008889
MonthOct -0.009271 0.2015 -0.04602 0.9633
MonthSep -0.01598 0.2123 -0.07526 0.94
ProductRelated_Duration 0.0001047 1.342e-05 7.796 6.395e-15
WeekendTRUE 0.1017 0.0708 1.437 0.1508 
global.measure <- function(m) {
cbind(
Deviance = m$deviance,
Null.Deviance = m$null.deviance,
AIC = m$aic,
DF = df.residual(m)
)
}

goodness <- rbind(
Full    = global.measure(full.model),
Reduced = global.measure(reduced.model),
Final   = global.measure(final.model)
)

pander(goodness, caption = "Global Goodness-of-Fit: Deviance & AIC")
Global Goodness-of-Fit: Deviance & AIC
Deviance Null.Deviance AIC DF
7167 10625 7205 12311
7551 10625 7563 12324
7169 10625 7203 12313 
coef_tab <- summary(final.model)$coef
odds <- exp(coef(final.model))
out <- cbind(coef_tab, `Odds Ratio` = odds)
pander(out, digits = 4, caption = "Final Model Coefficients with Odds Ratios")
Final Model Coefficients with Odds Ratios (continued below)
  Estimate Std. Error z value
(Intercept) -1.666 0.1729 -9.636
BounceRates -3.788 3.245 -1.168
ExitRates -16.17 2.336 -6.921
PageValues 0.08227 0.002399 34.29
VisitorTypeOther -0.5388 0.5252 -1.026
VisitorTypeReturning_Visitor -0.3099 0.08507 -3.643
MonthDec -0.6126 0.1817 -3.372
MonthFeb -1.812 0.6357 -2.85
MonthJul 0.08542 0.2182 0.3915
MonthJune -0.309 0.2752 -1.123
MonthMar -0.5232 0.1793 -2.917
MonthMay -0.5849 0.1692 -3.458
MonthNov 0.5389 0.1621 3.324
MonthOct -0.009271 0.2015 -0.04602
MonthSep -0.01598 0.2123 -0.07526
ProductRelated_Duration 0.0001047 1.342e-05 7.796
WeekendTRUE 0.1017 0.0708 1.437
  Pr(>|z|) Odds Ratio
(Intercept) 5.66e-22 0.189
BounceRates 0.243 0.02264
ExitRates 4.5e-12 9.504e-08
PageValues 1.14e-257 1.086
VisitorTypeOther 0.3049 0.5834
VisitorTypeReturning_Visitor 0.0002692 0.7335
MonthDec 0.0007462 0.5419
MonthFeb 0.004373 0.1634
MonthJul 0.6954 1.089
MonthJune 0.2614 0.7342
MonthMar 0.003533 0.5926
MonthMay 0.0005451 0.5572
MonthNov 0.0008889 1.714
MonthOct 0.9633 0.9908
MonthSep 0.94 0.9841
ProductRelated_Duration 6.395e-15 1
WeekendTRUE 0.1508 1.107 
phat <- fitted(final.model)


roc_obj <- roc(dat2$Revenue, phat)   
## Setting levels: control = No, case = Yes
## Setting direction: controls < cases
plot(roc_obj, main = "ROC Curve — Final Logistic Model")

auc_val <- auc(roc_obj)
auc_val
## Area under the curve: 0.8969
y_num <- as.numeric(dat2$Revenue) - 1
hoslem.test(y_num, phat, g = 10)
## 
##  Hosmer and Lemeshow goodness of fit (GOF) test
## 
## data:  y_num, phat
## X-squared = 180.45, df = 8, p-value < 2.2e-16