This notebook walks through a full exploratory data analysis (EDA) of
the store_sales.csv dataset using base R
only — no ggplot2, no dplyr, just the
tools that ship with R itself.
The goal is twofold. First, to give you a comprehensive reference for the kinds of plots you should be producing whenever you start working with a new dataset. Second, to build the habit of pairing every visualization with a short written interpretation. A plot without commentary is just decoration; a plot with a clear takeaway is analysis.
The structure of this notebook follows the natural progression of EDA: we start with one variable at a time (univariate), move on to relationships between two variables (bivariate), and finish with summary panels that bring everything together.
Reproducibility matters. Rather than scattering plots across your working directory, we will save every figure to a structured set of folders. Setting this up once at the top of the script saves a lot of time later.
folders <- c(
"plots/01_univariate_numeric",
"plots/02_univariate_categorical",
"plots/03_bivariate_numeric",
"plots/04_bivariate_numeric_x_categorical",
"plots/05_bivariate_categorical",
"plots/06_correlation",
"plots/07_student_dataset",
"plots/08_overview"
)
for (f in folders) {
dir.create(f, recursive = TRUE, showWarnings = FALSE)
}
cat("All output folders created under: plots/\n")#> All output folders created under: plots/This small wrapper around png() keeps our plotting code
clean. You can adjust width, height, and resolution depending on the
plot.
save_png <- function(path, width = 900, height = 650, res = 120) {
png(filename = path, width = width, height = height, res = res)
}Useful color references:
Before doing anything else, get a feel for the shape and content of your data.
Before plotting, it is often useful to slice the data and compute simple statistics on the resulting subsets. These exercises help you build intuition for what is in the data.
accessories_low_amount <- store_data[
store_data$Category == "Accessories" &
store_data$DiscountApplied > 15 &
store_data$Amount < 90,
]
nrow(accessories_low_amount)
head(accessories_low_amount)
mean(accessories_low_amount$Amount)
median(accessories_low_amount$Amount)
sd(accessories_low_amount$Amount)Here we compare the spending behavior of male and female customers using a simple summary table.
female_transactions <- store_data[store_data$Gender == "Female", ]
male_transactions <- store_data[store_data$Gender == "Male", ]
summary_table <- data.frame(
Gender = c("Female", "Male"),
Mean = c(mean(female_transactions$Amount, na.rm = TRUE),
mean(male_transactions$Amount, na.rm = TRUE)),
Median = c(median(female_transactions$Amount, na.rm = TRUE),
median(male_transactions$Amount, na.rm = TRUE)),
SD = c(sd(female_transactions$Amount, na.rm = TRUE),
sd(male_transactions$Amount, na.rm = TRUE))
)
summary_tableThis section covers plots for understanding a single numeric variable at a time. The standard toolkit includes histograms, density plots, boxplots, strip plots, ECDFs, and Q-Q plots.
Output folder:
plots/01_univariate_numeric/
A histogram divides the range of the variable into bins and counts
how many observations fall into each. The choice of breaks
matters: too few bins hides structure, too many adds noise.
save_png("plots/01_univariate_numeric/hist_amount.png")
hist(store_data$Amount,
main = "Distribution of Transaction Amount",
xlab = "Amount ($)",
col = "steelblue",
border = "white",
breaks = 20)
abline(v = mean(store_data$Amount),
col = "red", lty = 2, lwd = 2)
legend("topright",
legend = paste("Mean =", round(mean(store_data$Amount), 2)),
col = "red", lty = 2, lwd = 2)
dev.off()The dashed red line marks the mean. Adding the mean (and optionally the median) helps you see at a glance whether the distribution is skewed.
abline()quick reference:
ltycontrols line type (2 = dashed, 3 = dotted)lwdcontrols line thickness
save_png("plots/01_univariate_numeric/hist_age.png")
hist(store_data$Age,
main = "Age Distribution of Customers",
xlab = "Age (years)",
col = "darkorange",
border = "white",
breaks = 20)
dev.off()Practice exercise: Modify the chunk above to add both the mean and median as vertical lines, and include a legend explaining each. The full solution is shown below.
save_png("plots/01_univariate_numeric/hist_age.png")
hist(store_data$Age,
main = "Age Distribution of Customers",
xlab = "Age (years)",
col = "darkorange",
border = "white",
breaks = 10)
mean_age <- mean(store_data$Age, na.rm = TRUE)
median_age <- median(store_data$Age, na.rm = TRUE)
abline(v = mean_age, col = "red", lty = 2, lwd = 2)
abline(v = median_age, col = "blue", lty = 2, lwd = 2)
legend("topright",
legend = c(paste("Mean =", round(mean_age, 2)),
paste("Median =", round(median_age, 2))),
col = c("red", "blue"),
lty = 2,
lwd = 2)
dev.off()# Discount Applied
save_png("plots/01_univariate_numeric/hist_discount.png")
hist(store_data$DiscountApplied,
main = "Distribution of Discount Applied",
xlab = "Discount Applied",
col = "mediumpurple",
border = "white",
breaks = 20)
dev.off()
# Previous Purchases
save_png("plots/01_univariate_numeric/hist_previous_purchases.png")
hist(store_data$PreviousPurchases,
main = "Distribution of Previous Purchases",
xlab = "Number of Previous Purchases",
col = "forestgreen",
border = "white",
breaks = 20)
dev.off()A density plot is a smoothed version of a histogram. It often makes the shape of the distribution clearer, especially when comparing groups.
save_png("plots/01_univariate_numeric/density_amount.png")
d <- density(store_data$Amount)
plot(d,
main = "Density of Transaction Amount",
xlab = "Amount ($)",
col = "steelblue",
lwd = 2)
polygon(d, col = rgb(0.27, 0.51, 0.71, 0.25), border = NA)
abline(v = mean(store_data$Amount), col = "red", lty = 2, lwd = 2)
abline(v = median(store_data$Amount), col = "darkgreen", lty = 3, lwd = 2)
legend("topright",
legend = c("Mean", "Median"),
col = c("red", "darkgreen"),
lty = c(2, 3), lwd = 2)
dev.off()A common rule of thumb: any value below Q1 - 1.5 * IQR
or above Q3 + 1.5 * IQR is flagged as a potential outlier.
The function below returns the rows of the data frame that meet this
criterion.
get_outlier_rows <- function(df, column) {
x <- df[[column]]
Q1 <- quantile(x, 0.25, na.rm = TRUE)
Q3 <- quantile(x, 0.75, na.rm = TRUE)
IQR <- Q3 - Q1
lower <- Q1 - 1.5 * IQR
upper <- Q3 + 1.5 * IQR
df[x < lower | x > upper, ]
}
out_amount <- get_outlier_rows(store_data, "Amount")
nrow(out_amount)
colSums(is.na(store_data))The boxplot is one of the most efficient ways to see the median, the spread, and any extreme values in a single picture.
# Standard boxplot
save_png("plots/01_univariate_numeric/boxplot_amount.png")
boxplot(store_data$Amount,
main = "Boxplot of Transaction Amount",
ylab = "Amount ($)",
col = "lightblue",
border = "steelblue")
dev.off()
# Log-transformed boxplot — useful when the distribution is heavily skewed
save_png("plots/01_univariate_numeric/boxplot_log_amount.png")
boxplot(log(store_data$Amount),
main = "Boxplot of Transaction Log Amount",
ylab = "log(Amount)",
col = "lightblue",
border = "steelblue")
dev.off()save_png("plots/01_univariate_numeric/boxplot_amount_annotated.png")
boxplot(store_data$Amount,
main = "Amount — 5-Number Summary",
ylab = "Amount ($)",
col = "lightblue",
border = "steelblue")
five_num <- fivenum(store_data$Amount)
labels <- c("Min", "Q1", "Median", "Q3", "Max")
mtext(paste(labels, "=", round(five_num, 1)),
side = 4, at = five_num, las = 1, cex = 0.75, col = "steelblue")
dev.off()
cat("\n5-Number Summary of Amount:\n")
print(fivenum(store_data$Amount))save_png("plots/01_univariate_numeric/boxplot_age_annotated.png", width = 400)
boxplot(store_data$Age,
main = "Age — 5-Number Summary",
ylab = "Age (years)",
col = "lightblue",
border = "steelblue")
five_num <- fivenum(store_data$Age)
labels <- c("Min", "Q1", "Median", "Q3", "Max")
mtext(paste(labels, "=", round(five_num, 1)),
side = 4, at = five_num, las = 1, cex = 0.75, col = "steelblue")
dev.off()Strip plots show every individual observation. With jitter and transparency, they reveal density patterns that boxplots can hide.
save_png("plots/01_univariate_numeric/stripplot_amount.png")
stripchart(store_data$Amount,
method = "jitter",
pch = 19,
col = rgb(0.27, 0.51, 0.71, 0.15),
main = "Strip Plot of Transaction Amount",
xlab = "Amount ($)")
abline(v = mean(store_data$Amount), col = "red", lty = 2, lwd = 2)
abline(v = median(store_data$Amount), col = "darkgreen", lty = 3, lwd = 2)
legend("topright",
legend = c("Mean", "Median"),
col = c("red", "darkgreen"),
lty = c(2, 3), lwd = 2)
dev.off()The semi-transparent points (alpha = 0.15 in the RGB call) reveal density: darker regions indicate where many transactions cluster.
The ECDF answers the question: “what fraction of observations are at or below this value?” It is a useful complement to histograms and boxplots.
save_png("plots/01_univariate_numeric/ecdf_amount.png")
plot(ecdf(store_data$Amount),
main = "ECDF of Transaction Amount",
xlab = "Amount ($)",
ylab = "Cumulative Proportion",
col = "steelblue",
lwd = 2,
pch = NA)
abline(v = median(store_data$Amount), h = 0.5, col = "red", lty = 2)
legend("bottomright",
legend = paste("Median =", round(median(store_data$Amount), 2)),
col = "red", lty = 2)
dev.off()Q-Q plots compare the quantiles of your data against the quantiles of a theoretical distribution (here, the normal). If the points follow the reference line closely, the data is approximately normal.
# Q-Q plot for Amount
save_png("plots/01_univariate_numeric/qqplot_amount.png")
qqnorm(store_data$Amount,
main = "Normal Q-Q Plot of Transaction Amount",
pch = 19,
col = rgb(0.27, 0.51, 0.71, 0.4),
cex = 0.6)
qqline(store_data$Amount, col = "red", lwd = 2)
dev.off()
# Q-Q plot for Age
save_png("plots/01_univariate_numeric/qqplot_age.png")
qqnorm(store_data$Age,
main = "Normal Q-Q Plot of Customer Age",
pch = 19,
col = rgb(0.85, 0.33, 0.10, 0.4),
cex = 0.6)
qqline(store_data$Age, col = "red", lwd = 2)
dev.off()For categorical variables, the standard tools are bar charts (raw counts), pie charts (proportions), and Pareto charts (sorted bars with a cumulative line).
Output folder:
plots/02_univariate_categorical/
save_png("plots/02_univariate_categorical/bar_category.png", width = 1000)
cat_counts <- sort(table(store_data$Category), decreasing = TRUE)
barplot(cat_counts,
main = "Transactions by Product Category",
ylab = "Count",
col = "steelblue",
border = "white")
dev.off()# Season
save_png("plots/02_univariate_categorical/bar_season.png")
barplot(table(store_data$Season),
main = "Transactions by Season",
ylab = "Count",
xlab = "Season",
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "white")
dev.off()
# Gender
save_png("plots/02_univariate_categorical/bar_gender.png")
barplot(table(store_data$Gender),
main = "Transactions by Gender",
ylab = "Count",
col = c("tomato", "steelblue"),
border = "white")
dev.off()
# Payment Method
save_png("plots/02_univariate_categorical/bar_payment_method.png")
barplot(table(store_data$PaymentMethod),
main = "Transactions by Payment Method",
ylab = "Count",
col = c("steelblue", "darkorange"),
border = "white")
dev.off()Pie charts are best reserved for showing how a whole is divided among a small number of categories. With many categories, a sorted bar chart is almost always clearer.
# Category
save_png("plots/02_univariate_categorical/pie_category.png",
width = 900, height = 900)
cat_prop <- prop.table(table(store_data$Category))
pie(cat_prop,
main = "Share of Transactions by Category",
col = rainbow(length(cat_prop)),
labels = paste0(names(cat_prop), "\n",
round(cat_prop * 100, 1), "%"))
dev.off()
# Season
save_png("plots/02_univariate_categorical/pie_season.png",
width = 800, height = 800)
season_prop <- prop.table(table(store_data$Season))
pie(season_prop,
main = "Share of Transactions by Season",
col = c("darkorange", "steelblue", "gold", "skyblue"),
labels = paste0(names(season_prop), "\n",
round(season_prop * 100, 1), "%"))
dev.off()A Pareto chart sorts bars from largest to smallest and overlays a cumulative percentage line. It is the standard tool for asking: “which categories drive the majority of activity?”
save_png("plots/02_univariate_categorical/pareto_category.png", width = 1000)
cat_sorted <- sort(table(store_data$Category), decreasing = TRUE)
cat_cum_pct <- cumsum(cat_sorted) / sum(cat_sorted) * 100
par(mar = c(8, 4, 3, 4))
bp_pos <- barplot(cat_sorted,
main = "Pareto Chart - Product Category",
ylab = "Count",
col = "steelblue",
border = "white",
las = 2,
ylim = c(0, max(cat_sorted) * 1.1))
par(new = TRUE)
plot(bp_pos, cat_cum_pct,
type = "b", pch = 19, col = "red", lwd = 2,
axes = FALSE, xlab = "", ylab = "",
ylim = c(0, 100))
axis(side = 4, at = seq(0, 100, 20),
labels = paste0(seq(0, 100, 20), "%"), col.axis = "red")
mtext("Cumulative %", side = 4, line = 3, col = "red")
abline(h = 80, col = "red", lty = 2)
par(mar = c(5, 4, 4, 2))
dev.off()The horizontal line at 80% marks the classic Pareto threshold: which categories together account for 80% of transactions?
When comparing two numeric variables, the scatter plot is your starting point. Adding a regression line gives a quick read on linear association.
Output folder:
plots/03_bivariate_numeric/
save_png("plots/03_bivariate_numeric/scatter_amount_vs_discount.png")
plot(store_data$DiscountApplied, store_data$Amount,
main = "Amount vs Discount Applied",
xlab = "Discount Applied",
ylab = "Amount ($)",
pch = 19,
col = rgb(0.27, 0.51, 0.71, 0.3),
cex = 0.7)
abline(lm(Amount ~ DiscountApplied, data = store_data),
col = "red", lwd = 2)
dev.off()# Amount vs Previous Purchases
save_png("plots/03_bivariate_numeric/scatter_amount_vs_prev_purchases.png")
plot(store_data$PreviousPurchases, store_data$Amount,
main = "Amount vs Previous Purchases",
xlab = "Previous Purchases",
ylab = "Amount ($)",
pch = 19,
col = rgb(0.85, 0.33, 0.10, 0.3),
cex = 0.7)
abline(lm(Amount ~ PreviousPurchases, data = store_data),
col = "red", lwd = 2)
dev.off()
# Amount vs Age
save_png("plots/03_bivariate_numeric/scatter_amount_vs_age.png")
plot(store_data$Age, store_data$Amount,
main = "Amount vs Customer Age",
xlab = "Age (years)",
ylab = "Amount ($)",
pch = 19,
col = rgb(0.20, 0.63, 0.17, 0.3),
cex = 0.7)
abline(lm(Amount ~ Age, data = store_data),
col = "red", lwd = 2)
dev.off()
# Amount vs Item Rating
save_png("plots/03_bivariate_numeric/scatter_amount_vs_itemrating.png")
plot(store_data$ItemRating, store_data$Amount,
main = "Amount vs Item Rating",
xlab = "Item Rating (1-5)",
ylab = "Amount ($)",
pch = 19,
col = rgb(0.55, 0.20, 0.75, 0.25),
cex = 0.7)
abline(lm(Amount ~ ItemRating, data = store_data),
col = "red", lwd = 2)
dev.off()Adding a third variable as color in a scatter plot is one of the most effective ways to reveal subgroup differences.
save_png("plots/03_bivariate_numeric/scatter_amount_vs_age_by_gender.png")
gender_cols <- c("Female" = "tomato", "Male" = "steelblue")
plot(store_data$Age, store_data$Amount,
main = "Amount vs Age (by Gender)",
xlab = "Age (years)",
ylab = "Amount ($)",
pch = 19,
col = gender_cols[store_data$Gender],
cex = 0.7)
legend("topright",
legend = names(gender_cols),
col = gender_cols,
pch = 19)
dev.off()When you have several numeric variables, the pairs()
function produces a grid of all pairwise scatter plots in one go. This
is one of the fastest ways to scan for relationships.
num_vars <- store_data[, c("Age", "Amount", "PreviousPurchases", "ItemRating")]
save_png("plots/03_bivariate_numeric/scatter_matrix_all_numeric.png",
width = 1000, height = 1000)
pairs(num_vars,
main = "Scatter Plot Matrix - Numeric Variables",
pch = 19,
col = rgb(0.27, 0.51, 0.71, 0.2),
cex = 0.5,
panel = panel.smooth)
dev.off()When one variable is numeric and the other is categorical, side-by-side boxplots are the workhorse. They show how the distribution of the numeric variable differs across categories.
Output folder:
plots/04_bivariate_numeric_x_categorical/
# Amount by Season
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_amount_by_season.png")
boxplot(Amount ~ Season,
data = store_data,
main = "Transaction Amount by Season",
xlab = "Season",
ylab = "Amount ($)",
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "grey30")
dev.off()
# Amount by Gender
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_amount_by_gender.png")
boxplot(Amount ~ Gender,
data = store_data,
main = "Transaction Amount by Gender",
xlab = "Gender",
ylab = "Amount ($)",
col = c("tomato", "steelblue"),
border = "grey30")
dev.off()
# Amount by Category — wider plot, rotated labels
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_amount_by_category.png",
width = 1100)
par(mar = c(9, 4, 3, 1))
boxplot(Amount ~ Category,
data = store_data,
main = "Transaction Amount by Category",
xlab = "",
ylab = "Amount ($)",
col = rainbow(length(unique(store_data$Category)), alpha = 0.7),
border = "grey30",
las = 2)
par(mar = c(5, 4, 4, 2))
dev.off()
# Amount by Payment Method
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_amount_by_payment.png")
boxplot(Amount ~ PaymentMethod,
data = store_data,
main = "Transaction Amount by Payment Method",
xlab = "Payment Method",
ylab = "Amount ($)",
col = c("steelblue", "darkorange"),
border = "grey30")
dev.off()
# Item Rating by Season
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_itemrating_by_season.png")
boxplot(ItemRating ~ Season,
data = store_data,
main = "Item Rating by Season",
xlab = "Season",
ylab = "Item Rating (1-5)",
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "grey30")
dev.off()Combining a boxplot with a strip chart gives the best of both worlds: summary statistics plus the raw data.
save_png("plots/04_bivariate_numeric_x_categorical/boxplot_strip_amount_by_gender.png")
boxplot(Amount ~ Gender,
data = store_data,
main = "Amount by Gender (with raw data points)",
xlab = "Gender",
ylab = "Amount ($)",
col = c("mistyrose", "lightblue"),
border = "grey30")
stripchart(Amount ~ Gender,
data = store_data,
add = TRUE,
vertical = TRUE,
method = "jitter",
pch = 19,
col = c(rgb(0.8, 0.1, 0.1, 0.15),
rgb(0.1, 0.3, 0.7, 0.15)),
cex = 0.5)
dev.off()save_png("plots/04_bivariate_numeric_x_categorical/stripchart_amount_by_season.png")
stripchart(Amount ~ Season,
data = store_data,
method = "jitter",
vertical = TRUE,
pch = 19,
col = c(rgb(0.85, 0.41, 0.0, 0.3),
rgb(0.27, 0.51, 0.71, 0.3),
rgb(0.85, 0.75, 0.0, 0.3),
rgb(0.53, 0.81, 0.98, 0.3)),
main = "Amount by Season (Strip Chart)",
xlab = "Season",
ylab = "Amount ($)")
dev.off()When both variables are categorical, contingency tables and bar charts (side-by-side, stacked, or proportional) are your main tools. Mosaic plots add another perspective by encoding both counts and proportions in a single figure.
Output folder:
plots/05_bivariate_categorical/
# Side-by-side: Gender x Season
save_png("plots/05_bivariate_categorical/bar_sidebyside_gender_x_season.png")
barplot(gs_table,
beside = TRUE,
main = "Transactions: Gender x Season",
xlab = "Season",
ylab = "Count",
col = c("tomato", "steelblue"),
border = "white",
legend.text = rownames(gs_table),
args.legend = list(x = "topright", bty = "n"))
dev.off()
# Stacked: Gender x Season
save_png("plots/05_bivariate_categorical/bar_stacked_gender_x_season.png")
barplot(gs_table,
beside = FALSE,
main = "Stacked: Gender x Season",
xlab = "Season",
ylab = "Count",
col = c("tomato", "steelblue"),
border = "white",
legend.text = rownames(gs_table),
args.legend = list(x = "topright", bty = "n"))
dev.off()
# Proportional stacked: Gender x Season
save_png("plots/05_bivariate_categorical/bar_proportional_gender_x_season.png")
gs_prop <- prop.table(gs_table, margin = 2)
barplot(gs_prop,
beside = FALSE,
main = "Proportional: Gender x Season",
xlab = "Season",
ylab = "Proportion",
col = c("tomato", "steelblue"),
border = "white",
ylim = c(0, 1),
legend.text = rownames(gs_table),
args.legend = list(x = "topright", bty = "n"))
dev.off()
# Side-by-side: Gender x Payment Method
save_png("plots/05_bivariate_categorical/bar_sidebyside_gender_x_payment.png")
barplot(gp_table,
beside = TRUE,
main = "Transactions: Gender x Payment Method",
xlab = "Payment Method",
ylab = "Count",
col = c("tomato", "steelblue"),
border = "white",
legend.text = rownames(gp_table),
args.legend = list(x = "topright", bty = "n"))
dev.off()
# Side-by-side: Category x Season
save_png("plots/05_bivariate_categorical/bar_sidebyside_category_x_season.png",
width = 1200)
par(mar = c(5, 4, 3, 12))
barplot(cs_table,
beside = TRUE,
main = "Transactions: Category x Season",
xlab = "Season",
ylab = "Count",
col = rainbow(nrow(cs_table), alpha = 0.8),
border = "white",
legend.text = rownames(cs_table),
args.legend = list(x = "topright", xpd = TRUE,
bty = "n", inset = c(-0.25, 0)))
par(mar = c(5, 4, 4, 2))
dev.off()# Gender x Payment Method
save_png("plots/05_bivariate_categorical/mosaic_gender_x_payment.png")
mosaicplot(gp_table,
main = "Gender x Payment Method",
xlab = "Gender",
ylab = "Payment Method",
col = c("steelblue", "tomato"),
border = "white")
dev.off()
# Category x Season
save_png("plots/05_bivariate_categorical/mosaic_category_x_season.png",
width = 1000)
par(mar = c(5, 9, 3, 1))
mosaicplot(cs_table,
main = "Category x Season",
col = rainbow(4, alpha = 0.7),
border = "white",
las = 1)
par(mar = c(5, 4, 4, 2))
dev.off()When you want to test whether two categorical variables are
statistically independent, use chisq.test().
cat("\n--- Contingency Table: Gender x PaymentMethod ---\n")
print(gp_table)
cat("\nRow proportions:\n")
print(round(prop.table(gp_table, margin = 1), 3))
cat("\nChi-square test:\n")
print(chisq.test(gp_table))Correlation and covariance summarize linear relationships between numeric variables. Correlation is unitless and easier to interpret; covariance retains the units of the original variables.
Output folder:
plots/06_correlation/
This is a heatmap built entirely from base R — no extra packages required.
save_png("plots/06_correlation/heatmap_correlation.png",
width = 800, height = 750)
cor_colors <- colorRampPalette(c("tomato", "white", "steelblue"))(200)
par(mar = c(6, 6, 3, 2))
image(1:ncol(cor_matrix), 1:nrow(cor_matrix),
t(cor_matrix)[, nrow(cor_matrix):1],
col = cor_colors,
xlab = "", ylab = "",
axes = FALSE,
main = "Correlation Heatmap - Numeric Variables",
zlim = c(-1, 1))
axis(1, at = 1:ncol(cor_matrix),
labels = colnames(cor_matrix), las = 2, cex.axis = 0.85)
axis(2, at = nrow(cor_matrix):1,
labels = colnames(cor_matrix), las = 1, cex.axis = 0.85)
for (i in 1:nrow(cor_matrix)) {
for (j in 1:ncol(cor_matrix)) {
text(j, nrow(cor_matrix) + 1 - i,
labels = round(cor_matrix[i, j], 2),
cex = 0.9, font = 2,
col = ifelse(abs(cor_matrix[i, j]) > 0.5, "white", "black"))
}
}
par(mar = c(5, 4, 4, 2))
dev.off()corrplotIf you have the corrplot package installed, it produces
nicer correlation visualizations with much less code.
if (requireNamespace("corrplot", quietly = TRUE)) {
library(corrplot)
# Full matrix
save_png("plots/06_correlation/corrplot_full.png",
width = 800, height = 750)
corrplot(cor_matrix, method = "color", type = "full",
addCoef.col = "black", tl.col = "black", tl.srt = 45,
title = "Correlation Matrix", mar = c(0, 0, 1, 0))
dev.off()
# Upper triangle only
save_png("plots/06_correlation/corrplot_upper.png",
width = 800, height = 750)
corrplot(cor_matrix, method = "color", type = "upper",
addCoef.col = "black", tl.col = "black", tl.srt = 45,
title = "Correlation Matrix (Upper Triangle)",
mar = c(0, 0, 1, 0))
dev.off()
}To consolidate what we have covered, here is a small, easy-to-read example using a hand-crafted student dataset. Working with a tiny dataset like this makes it easy to verify what each plot is showing.
Output folder:
plots/07_student_dataset/
student_data <- data.frame(
StudentID = 1:10,
StudyHours = c(2, 4, 3, 5, 6, 7, 4, 8, 5, 6),
Attendance = c(50, 70, 60, 80, 85, 90, 75, 95, 80, 85),
Score = c(55, 72, 63, 78, 84, 88, 74, 92, 79, 85)
)
cat("\n===== Student Dataset =====\n")
print(student_data)
s_cov <- cov(student_data[, c("StudyHours", "Attendance", "Score")])
s_cor <- cor(student_data[, c("StudyHours", "Attendance", "Score")])
cat("\n--- Covariance Matrix ---\n"); print(round(s_cov, 4))
cat("\n--- Correlation Matrix ---\n"); print(round(s_cor, 3))save_png("plots/07_student_dataset/scatter_score_vs_studyhours.png")
plot(student_data$StudyHours, student_data$Score,
main = "Score vs Study Hours",
xlab = "Study Hours",
ylab = "Score",
pch = 19, col = "steelblue", cex = 1.5)
abline(lm(Score ~ StudyHours, data = student_data),
col = "red", lwd = 2)
text(student_data$StudyHours, student_data$Score,
labels = student_data$StudentID,
pos = 3, cex = 0.8, col = "grey40")
dev.off()save_png("plots/07_student_dataset/scatter_score_vs_attendance.png")
plot(student_data$Attendance, student_data$Score,
main = "Score vs Attendance (%)",
xlab = "Attendance (%)",
ylab = "Score",
pch = 19, col = "darkorange", cex = 1.5)
abline(lm(Score ~ Attendance, data = student_data),
col = "red", lwd = 2)
dev.off()save_png("plots/07_student_dataset/heatmap_correlation_student.png",
width = 700, height = 650)
cor_colors <- colorRampPalette(c("tomato", "white", "steelblue"))(200)
par(mar = c(5, 5, 3, 2))
image(1:3, 1:3,
t(s_cor)[, 3:1],
col = cor_colors,
xlab = "", ylab = "",
axes = FALSE,
main = "Correlation Heatmap - Student Dataset",
zlim = c(-1, 1))
axis(1, at = 1:3, labels = colnames(s_cor), las = 2)
axis(2, at = 3:1, labels = colnames(s_cor), las = 1)
for (i in 1:3) {
for (j in 1:3) {
text(j, 4 - i, round(s_cor[i, j], 3),
cex = 1.1, font = 2,
col = ifelse(abs(s_cor[i, j]) > 0.5, "white", "black"))
}
}
par(mar = c(5, 4, 4, 2))
dev.off()ct_example <- matrix(
c(15, 2, 5, 8),
nrow = 2,
dimnames = list(Attendance = c("High", "Low"),
Result = c("Pass", "Fail"))
)
cat("\n--- Contingency Table: Attendance x Result ---\n")
print(ct_example)
cat("\nRow %:\n"); print(round(prop.table(ct_example, 1) * 100, 1))
cat("\nColumn %:\n"); print(round(prop.table(ct_example, 2) * 100, 1))
cat("\nChi-square:\n"); print(chisq.test(ct_example))
save_png("plots/07_student_dataset/mosaic_attendance_x_result.png")
mosaicplot(ct_example,
main = "Attendance x Pass/Fail",
col = c("steelblue", "tomato"),
border = "white")
dev.off()Once you have explored the data thoroughly, it is often useful to
assemble a single summary figure that captures the key findings. The
par(mfrow = c(rows, cols)) call allows you to combine
multiple plots in a grid.
Output folder:
plots/08_overview/
save_png("plots/08_overview/overview_6panel.png",
width = 1400, height = 950, res = 130)
par(mfrow = c(2, 3))
# Panel 1: Distribution of Amount
hist(store_data$Amount,
col = "steelblue", border = "white",
main = "Distribution of Amount",
xlab = "Amount ($)", breaks = 25)
abline(v = mean(store_data$Amount), col = "red", lty = 2, lwd = 2)
# Panel 2: Age distribution
hist(store_data$Age,
col = "darkorange", border = "white",
main = "Age Distribution", xlab = "Age (years)")
# Panel 3: Amount by Season
boxplot(Amount ~ Season, data = store_data,
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "grey30",
main = "Amount by Season",
xlab = "Season", ylab = "Amount ($)", cex.axis = 0.8)
# Panel 4: Amount vs Age
plot(store_data$Age, store_data$Amount,
pch = 19, col = rgb(0.27, 0.51, 0.71, 0.3), cex = 0.6,
main = "Amount vs Age",
xlab = "Age", ylab = "Amount ($)")
abline(lm(Amount ~ Age, data = store_data), col = "red", lwd = 2)
# Panel 5: Category counts
barplot(sort(table(store_data$Category), decreasing = TRUE),
col = "steelblue", border = "white",
main = "Category Counts", las = 2, cex.names = 0.6)
# Panel 6: ECDF of Amount
plot(ecdf(store_data$Amount),
col = "steelblue", lwd = 2, pch = NA,
main = "ECDF - Amount",
xlab = "Amount ($)", ylab = "Cumulative Proportion")
abline(v = median(store_data$Amount), h = 0.5, col = "red", lty = 2)
par(mfrow = c(1, 1))
dev.off()save_png("plots/08_overview/overview_univariate_numeric.png",
width = 1200, height = 900, res = 120)
par(mfrow = c(2, 2))
hist(store_data$Amount,
col = "steelblue", border = "white",
main = "Histogram: Amount", xlab = "Amount ($)", breaks = 25)
d <- density(store_data$Amount)
plot(d, col = "steelblue", lwd = 2,
main = "Density: Amount", xlab = "Amount ($)")
polygon(d, col = rgb(0.27, 0.51, 0.71, 0.2), border = NA)
boxplot(store_data$Amount,
col = "lightblue", border = "steelblue",
main = "Boxplot: Amount", ylab = "Amount ($)")
qqnorm(store_data$Amount,
main = "Q-Q Plot: Amount",
pch = 19, col = rgb(0.27, 0.51, 0.71, 0.4), cex = 0.6)
qqline(store_data$Amount, col = "red", lwd = 2)
par(mfrow = c(1, 1))
dev.off()save_png("plots/08_overview/overview_bivariate_boxplots.png",
width = 1200, height = 900, res = 120)
par(mfrow = c(2, 2))
boxplot(Amount ~ Season, data = store_data,
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "grey30",
main = "Amount by Season",
xlab = "Season", ylab = "Amount ($)")
boxplot(Amount ~ Gender, data = store_data,
col = c("tomato", "steelblue"), border = "grey30",
main = "Amount by Gender",
xlab = "Gender", ylab = "Amount ($)")
boxplot(Amount ~ PaymentMethod, data = store_data,
col = c("steelblue", "darkorange"), border = "grey30",
main = "Amount by Payment Method",
xlab = "Payment Method", ylab = "Amount ($)")
boxplot(ItemRating ~ Season, data = store_data,
col = c("darkorange", "steelblue", "gold", "skyblue"),
border = "grey30",
main = "Item Rating by Season",
xlab = "Season", ylab = "Item Rating (1-5)")
par(mfrow = c(1, 1))
dev.off()A final convenience: print a tree-like summary of every figure saved during the session. Useful as a sanity check that nothing was missed.
cat("All plots saved. Folder structure:\n\n")
for (f in folders) {
files <- list.files(f)
cat(sprintf(" %s/ (%d files)\n", f, length(files)))
for (file in files) {
cat(sprintf(" - %s\n", file))
}
}This notebook covered the standard EDA toolkit in base R: histograms, density plots, boxplots, strip plots, ECDFs, and Q-Q plots for univariate exploration; bar charts, pie charts, and Pareto charts for categorical data; scatter plots, correlation heatmaps, and pairs plots for relationships among numeric variables; and side-by-side boxplots, contingency tables, and mosaic plots for mixed and categorical pairings.
Two takeaways worth keeping in mind as you build on this material:
First, every plot should answer a question. Before producing a figure, ask yourself what you are trying to learn from it. After producing it, write down what you actually learned. If you cannot, the plot is probably not earning its place in your analysis.
Second, base R is more than enough for serious EDA.
Specialized packages such as ggplot2 are powerful and worth
learning, but most of what you need can be done with the tools
demonstrated here. Mastering the basics first will make the advanced
packages much easier to pick up later.
Practice exercises:
- Reproduce the histogram of
Agewith both mean and median lines and a legend.- Create a side-by-side boxplot of
DiscountAppliedbyCategorywith rotated x-axis labels.- Build a 2x2 panel showing the four numeric variables, each with its own ECDF.
- Use
chisq.test()to test independence betweenGenderandCategory. Interpret the result in plain language.