Mini Project 1 — Event-Based Modeling of Bank Customer Visits

Marketing Analytics (MiM811) — Mini Project Report

1 Project Goal and Data Overview

1.1 Goal

This project aims to understand and predict daily bank customer visits (CUST) using:

• A time trend (TimeIndex)
• Weekday effects (DAYCAT)
• Special events (Payday, holidays, before/after holidays, etc.)
• Extra events (e.g., Christmas, New Year, Black Friday) and interactions

1.2 Load the dataset (Excel)

# # Load packages
library(tidyverse)   # # data wrangling + ggplot2
library(readxl)      # # read Excel
library(lubridate)   # # date functions (month(), wday(), etc.)
library(janitor)     # # clean_names()
library(broom)       # # tidy model outputs
library(knitr)       # # kable tables
library(kableExtra)  # # nicer tables
library(see)
# =========================

# # 1A) LOAD DATA

# =========================

# # Read Excel file (make sure it's in the same folder as this .Rmd)

raw <- read_excel("MiM811Quiz4MiniProj1Data.xlsx")

# # Clean column names (removes weird spaces like "  CUST")

data <- raw %>%
janitor::clean_names()

# # Quick look

glimpse(data)

## Rows: 254
## Columns: 13
## $ cust     <dbl> 1825, 1257, 969, 1672, 1098, 691, 672, 754, 972, 816, 717, 72…
## $ daycat   <chr> "Tuesday", "Wednesday", "Thursday", "Friday", "Monday", "Tues…
## $ date     <dttm> 2007-01-02, 2007-01-03, 2007-01-04, 2007-01-05, 2007-01-08, …
## $ month    <dbl> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1…
## $ daymon   <dbl> 2, 3, 4, 5, 8, 9, 10, 11, 12, 15, 16, 17, 18, 19, 22, 23, 24,…
## $ dayweek  <dbl> 2, 3, 4, 5, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 1, 2…
## $ special  <chr> "SP,FAC,AH", "0", "0", "SP", "0", "0", "0", "0", "0", "0", "0…
## $ payday   <dbl> 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0…
## $ sp       <dbl> 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0…
## $ fac      <dbl> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ holidays <dbl> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ bh       <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…
## $ ah       <dbl> 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0…

1.3 Quick interactive view

DT::datatable(
head(data, 20),
options = list(pageLength = 10, scrollX = TRUE),
caption = "First 20 rows of the dataset (interactive table)"
)

2 Part 1 — Clean and Prepare Data

2.1 Convert DATE + create TimeIndex

# =========================

# # 1B) FIX TYPES + CREATE TIME INDEX

# =========================

data <- data %>%
mutate(
# # Convert date to Date class (even if it already looks like a date)
date = as.Date(date),

# # Time trend index: 1, 2, 3, ..., n
time_index = row_number(),

# # DAYCAT as ordered factor (important for weekday comparisons)
daycat = factor(
  daycat,
  levels = c("Monday","Tuesday","Wednesday","Thursday","Friday","Saturday","Sunday"),
  ordered = TRUE
),

# # Month as a factor (optional but useful)
month_f = factor(month, levels = 1:12, labels = month.name)
)

# # Review structure + summary (required “review structure” idea)

str(data)

## tibble [254 × 15] (S3: tbl_df/tbl/data.frame)
##  $ cust      : num [1:254] 1825 1257 969 1672 1098 ...
##  $ daycat    : Ord.factor w/ 7 levels "Monday"<"Tuesday"<..: 2 3 4 5 1 2 3 4 5 1 ...
##  $ date      : Date[1:254], format: "2007-01-02" "2007-01-03" ...
##  $ month     : num [1:254] 1 1 1 1 1 1 1 1 1 1 ...
##  $ daymon    : num [1:254] 2 3 4 5 8 9 10 11 12 15 ...
##  $ dayweek   : num [1:254] 2 3 4 5 1 2 3 4 5 1 ...
##  $ special   : chr [1:254] "SP,FAC,AH" "0" "0" "SP" ...
##  $ payday    : num [1:254] 1 0 0 1 0 0 0 0 0 0 ...
##  $ sp        : num [1:254] 1 0 0 1 0 0 0 0 0 0 ...
##  $ fac       : num [1:254] 1 0 0 0 0 0 0 0 0 0 ...
##  $ holidays  : num [1:254] 1 0 0 0 0 0 0 0 0 0 ...
##  $ bh        : num [1:254] 0 0 0 0 0 0 0 0 0 0 ...
##  $ ah        : num [1:254] 1 0 0 0 0 0 0 0 0 0 ...
##  $ time_index: int [1:254] 1 2 3 4 5 6 7 8 9 10 ...
##  $ month_f   : Factor w/ 12 levels "January","February",..: 1 1 1 1 1 1 1 1 1 1 ...

summary(data)

##       cust              daycat        date                month       
##  Min.   : 404.0   Monday   :50   Min.   :2007-01-02   Min.   : 1.000  
##  1st Qu.: 785.8   Tuesday  :51   1st Qu.:2007-03-30   1st Qu.: 3.250  
##  Median : 930.5   Wednesday:50   Median :2007-06-30   Median : 6.500  
##  Mean   :1037.5   Thursday :51   Mean   :2007-06-30   Mean   : 6.476  
##  3rd Qu.:1183.5   Friday   :52   3rd Qu.:2007-09-30   3rd Qu.: 9.750  
##  Max.   :2068.0   Saturday : 0   Max.   :2007-12-31   Max.   :12.000  
##                   Sunday   : 0                                        
##      daymon         dayweek        special              payday     
##  Min.   : 1.00   Min.   :1.000   Length:254         Min.   :0.000  
##  1st Qu.: 8.00   1st Qu.:2.000   Class :character   1st Qu.:0.000  
##  Median :16.00   Median :3.000   Mode  :character   Median :0.000  
##  Mean   :15.83   Mean   :3.016                      Mean   :0.122  
##  3rd Qu.:23.00   3rd Qu.:4.000                      3rd Qu.:0.000  
##  Max.   :31.00   Max.   :5.000                      Max.   :1.000  
##                                                                    
##        sp               fac             holidays             bh         
##  Min.   :0.00000   Min.   :0.00000   Min.   :0.00000   Min.   :0.00000  
##  1st Qu.:0.00000   1st Qu.:0.00000   1st Qu.:0.00000   1st Qu.:0.00000  
##  Median :0.00000   Median :0.00000   Median :0.00000   Median :0.00000  
##  Mean   :0.09843   Mean   :0.03937   Mean   :0.04724   Mean   :0.01969  
##  3rd Qu.:0.00000   3rd Qu.:0.00000   3rd Qu.:0.00000   3rd Qu.:0.00000  
##  Max.   :1.00000   Max.   :1.00000   Max.   :1.00000   Max.   :1.00000  
##                                                                         
##        ah            time_index        month_f   
##  Min.   :0.00000   Min.   :  1.00   August : 23  
##  1st Qu.:0.00000   1st Qu.: 64.25   October: 23  
##  Median :0.00000   Median :127.50   January: 22  
##  Mean   :0.02756   Mean   :127.50   March  : 22  
##  3rd Qu.:0.00000   3rd Qu.:190.75   May    : 22  
##  Max.   :1.00000   Max.   :254.00   April  : 21  
##                                     (Other):121

# =========================

2.2 Understand the event flags (sanity checks)

# # 1C) UNDERSTAND EVENT FLAGS

# =========================

# # These event columns are your “special events” predictors.

# # Based on names, they typically mean:

# # payday   = payday indicator (1/0)

# # sp       = staff payday (1/0)  (guess from abbreviation)

# # fac      = faculty payday (1/0)

# # holidays = holiday day (1/0)

# # bh       = before-holiday (1/0)

# # ah       = after-holiday (1/0)

# # If your class notes define them differently, use the official meaning.

# # Count how many days are flagged as 1 (quick sanity check)

data %>%
summarise(across(c(payday, sp, fac, holidays, bh, ah), ~sum(.x, na.rm = TRUE))) %>%
kable(caption = "How many days are flagged for each event?") %>%
kableExtra::kable_styling(full_width = FALSE)

How many days are flagged for each event?

payday	sp	fac	holidays	bh	ah
31	25	10	12	5	7

# =========================

2.3 Create extra event flags (New Year, Christmas, Black Friday)

# =========================================

# # 2.3) OPTIONAL EXTRA EVENTS

# =========================================

data <- data %>%
mutate(
year = year(date),

# # New Year (Jan 1)
is_new_year = if_else(month(date) == 1 & day(date) == 1, 1, 0),

# # Christmas (Dec 25)
is_christmas = if_else(month(date) == 12 & day(date) == 25, 1, 0)


)

# # Black Friday approximation:

# # US definition: 4th Friday of November.

black_fridays <- data %>%
filter(month(date) == 11, wday(date, label = TRUE) == "Fri") %>%
group_by(year) %>%
arrange(date) %>%
mutate(friday_number = row_number()) %>%
filter(friday_number == 4) %>%
pull(date)

data <- data %>%
mutate(is_black_friday = if_else(date %in% black_fridays, 1, 0))

data %>%
filter(is_new_year == 1 | is_christmas == 1 | is_black_friday == 1) %>%
select(date, is_new_year, is_christmas, is_black_friday) %>%
arrange(date) %>%
  kable(caption = "Detected dates for optional extra events")

Detected dates for optional extra events

date	is_new_year	is_christmas	is_black_friday
2007-11-23	0	0	1

3 Part 2 — Visualize Customer Behavior

3.1 Daily visits over time (time series)

ggplot(data, aes(x = date, y = cust)) +
geom_line() +
labs(
title = "Daily Customer Visits Over Time",
x = "Date",
y = "CUST (Visits)"
)

3.2 Visits by weekday (DAYCAT)

ggplot(data, aes(x = daycat, y = cust)) +
geom_boxplot() +
labs(
title = "Customer Visits by Day of Week (DAYCAT)",
x = "Day of Week",
y = "CUST (Visits)"
)

3.3 Color by payday and facet by holidays

plot_data <- data %>%
mutate(
payday_label = factor(payday, levels = c(0,1), labels = c("Not Payday","Payday")),
holiday_label = factor(holidays, levels = c(0,1), labels = c("Not Holiday","Holiday"))
)

ggplot(plot_data, aes(x = date, y = cust, color = payday_label)) +
geom_line() +
facet_wrap(~ holiday_label, ncol = 1) +
labs(
title = "Daily Visits — Colored by Payday, Faceted by Holiday",
x = "Date",
y = "CUST (Visits)",
color = "Payday?"
)

3.4 Mark special dates with vertical lines

event_dates <- data %>%
filter(is_new_year == 1 | is_christmas == 1 | is_black_friday == 1) %>%
select(date) %>%
distinct()

ggplot(data, aes(x = date, y = cust)) +
geom_line() +
geom_vline(
data = event_dates,
aes(xintercept = as.numeric(date)),
linetype = "dashed"
) +
labs(
title = "Daily Visits with Special Event Markers (dashed lines)",
x = "Date",
y = "CUST (Visits)"
)

4 Part 3 — Regression Models

4.1 Model 1: baseline (trend + weekday)

m1 <- lm(cust ~ time_index + daycat, data = data)
summary(m1)

## 
## Call:
## lm(formula = cust ~ time_index + daycat, data = data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -572.91 -144.51  -35.06   92.18  962.35 
## 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept) 1.024e+03  2.924e+01  35.008  < 2e-16 ***
## time_index  8.908e-02  1.988e-01   0.448 0.654551    
## daycat.L    2.605e+02  3.253e+01   8.009 4.51e-14 ***
## daycat.Q    4.747e+02  3.263e+01  14.551  < 2e-16 ***
## daycat.C    1.205e+02  3.252e+01   3.706 0.000259 ***
## daycat^4    5.887e+01  3.269e+01   1.801 0.072949 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 232.3 on 248 degrees of freedom
## Multiple R-squared:  0.5456, Adjusted R-squared:  0.5365 
## F-statistic: 59.56 on 5 and 248 DF,  p-value: < 2.2e-16

kable(broom::tidy(m1), caption = "Model 1 coefficients (tidy)")

Model 1 coefficients (tidy)

term	estimate	std.error	statistic	p.value
(Intercept)	1023.7204995	29.2426951	35.0077343	0.0000000
time_index	0.0890758	0.1988343	0.4479902	0.6545514
daycat.L	260.4968532	32.5268084	8.0086816	0.0000000
daycat.Q	474.7217560	32.6256270	14.5505788	0.0000000
daycat.C	120.5438976	32.5248257	3.7062120	0.0002595
daycat^4	58.8745043	32.6934556	1.8008040	0.0729487

4.2 Model 2: add event flags

m2 <- lm(cust ~ time_index + daycat + payday + holidays + sp + fac + bh + ah, data = data)
summary(m2)

## 
## Call:
## lm(formula = cust ~ time_index + daycat + payday + holidays + 
##     sp + fac + bh + ah, data = data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -484.93 -123.98  -12.84   93.32  807.98 
## 
## Coefficients: (1 not defined because of singularities)
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept) 956.34903   23.46298  40.760  < 2e-16 ***
## time_index    0.06242    0.15627   0.399 0.689914    
## daycat.L    127.06006   28.86490   4.402 1.61e-05 ***
## daycat.Q    363.95776   27.94200  13.025  < 2e-16 ***
## daycat.C     47.74734   26.40794   1.808 0.071832 .  
## daycat^4     57.95678   25.57645   2.266 0.024330 *  
## payday      443.14322  128.13516   3.458 0.000641 ***
## holidays    302.35215   70.83869   4.268 2.83e-05 ***
## sp            0.97173  122.05212   0.008 0.993654    
## fac          82.61443  103.05425   0.802 0.423533    
## bh            6.55590  105.86856   0.062 0.950674    
## ah                 NA         NA      NA       NA    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 179.7 on 243 degrees of freedom
## Multiple R-squared:  0.7335, Adjusted R-squared:  0.7226 
## F-statistic:  66.9 on 10 and 243 DF,  p-value: < 2.2e-16

kable(broom::tidy(m2), caption = "Model 2 coefficients (tidy)")

Model 2 coefficients (tidy)

term	estimate	std.error	statistic	p.value
(Intercept)	956.3490313	23.4629831	40.7599080	0.0000000
time_index	0.0624229	0.1562731	0.3994474	0.6899144
daycat.L	127.0600608	28.8649022	4.4018878	0.0000161
daycat.Q	363.9577594	27.9419969	13.0254742	0.0000000
daycat.C	47.7473440	26.4079355	1.8080680	0.0718322
daycat^4	57.9567812	25.5764467	2.2660216	0.0243304
payday	443.1432183	128.1351649	3.4584044	0.0006415
holidays	302.3521510	70.8386880	4.2681783	0.0000283
sp	0.9717298	122.0521181	0.0079616	0.9936542
fac	82.6144252	103.0542468	0.8016596	0.4235331
bh	6.5558990	105.8685622	0.0619249	0.9506735
ah	NA	NA	NA	NA

4.3 Model 3: add month + interactions

m3 <- lm(
cust ~ time_index + month_f + daycat * payday + holidays * fac + sp + bh + ah,
data = data
)
summary(m3)

## 
## Call:
## lm(formula = cust ~ time_index + month_f + daycat * payday + 
##     holidays * fac + sp + bh + ah, data = data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -405.80  -89.02  -15.41   68.44  546.85 
## 
## Coefficients: (2 not defined because of singularities)
##                  Estimate Std. Error t value Pr(>|t|)    
## (Intercept)      1003.131     37.660  26.636  < 2e-16 ***
## time_index        -12.032      1.596  -7.538 1.12e-12 ***
## month_fFebruary   277.368     57.165   4.852 2.26e-06 ***
## month_fMarch      576.918     80.713   7.148 1.18e-11 ***
## month_fApril      867.141    110.434   7.852 1.59e-13 ***
## month_fMay       1213.237    142.457   8.517 2.24e-15 ***
## month_fJune      1511.428    174.634   8.655 9.03e-16 ***
## month_fJuly      1734.265    206.059   8.416 4.31e-15 ***
## month_fAugust    1925.714    239.655   8.035 5.00e-14 ***
## month_fSeptember 2051.795    273.620   7.499 1.42e-12 ***
## month_fOctober   2339.013    307.425   7.608 7.25e-13 ***
## month_fNovember  2604.374    341.224   7.632 6.25e-13 ***
## month_fDecember  2940.767    373.585   7.872 1.41e-13 ***
## daycat.L          140.937     24.401   5.776 2.49e-08 ***
## daycat.Q          362.260     23.462  15.440  < 2e-16 ***
## daycat.C           47.703     22.134   2.155  0.03220 *  
## daycat^4           68.904     21.205   3.249  0.00133 ** 
## payday            375.837    130.811   2.873  0.00445 ** 
## holidays          297.878     61.687   4.829 2.52e-06 ***
## fac               115.894    111.898   1.036  0.30143    
## sp                 56.356    170.155   0.331  0.74080    
## bh                 15.484     91.287   0.170  0.86546    
## ah                     NA         NA      NA       NA    
## daycat.L:payday   -26.654    133.276  -0.200  0.84166    
## daycat.Q:payday   -19.789    150.232  -0.132  0.89532    
## daycat.C:payday    77.822    129.226   0.602  0.54763    
## daycat^4:payday        NA         NA      NA       NA    
## holidays:fac      -34.927    180.755  -0.193  0.84695    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 147.6 on 228 degrees of freedom
## Multiple R-squared:  0.8313, Adjusted R-squared:  0.8128 
## F-statistic: 44.93 on 25 and 228 DF,  p-value: < 2.2e-16

kable(broom::tidy(m3), caption = "Model 3 coefficients (tidy)")

Model 3 coefficients (tidy)

term	estimate	std.error	statistic	p.value
(Intercept)	1003.13099	37.660147	26.6364063	0.0000000
time_index	-12.03244	1.596212	-7.5381213	0.0000000
month_fFebruary	277.36798	57.164786	4.8520777	0.0000023
month_fMarch	576.91819	80.712549	7.1478128	0.0000000
month_fApril	867.14071	110.433875	7.8521261	0.0000000
month_fMay	1213.23683	142.456517	8.5165414	0.0000000
month_fJune	1511.42795	174.634334	8.6548156	0.0000000
month_fJuly	1734.26505	206.058605	8.4163680	0.0000000
month_fAugust	1925.71396	239.654723	8.0353683	0.0000000
month_fSeptember	2051.79494	273.619898	7.4987051	0.0000000
month_fOctober	2339.01257	307.425347	7.6083921	0.0000000
month_fNovember	2604.37428	341.223612	7.6324562	0.0000000
month_fDecember	2940.76705	373.585047	7.8717472	0.0000000
daycat.L	140.93724	24.401140	5.7758464	0.0000000
daycat.Q	362.26042	23.462429	15.4400223	0.0000000
daycat.C	47.70280	22.133998	2.1551824	0.0321951
daycat^4	68.90354	21.205033	3.2493958	0.0013311
payday	375.83676	130.811376	2.8731199	0.0044481
holidays	297.87763	61.687237	4.8288373	0.0000025
fac	115.89402	111.898080	1.0357105	0.3014344
sp	56.35569	170.155324	0.3312015	0.7407965
bh	15.48419	91.286798	0.1696213	0.8654583
ah	NA	NA	NA	NA
daycat.L:payday	-26.65448	133.276496	-0.1999938	0.8416637
daycat.Q:payday	-19.78862	150.232377	-0.1317201	0.8953219
daycat.C:payday	77.82195	129.226210	0.6022149	0.5476295
daycat^4:payday	NA	NA	NA	NA
holidays:fac	-34.92723	180.754683	-0.1932300	0.8469509

4.4 Compare models (R² and Adjusted R²)

model_compare <- tibble(
model = c("m1: trend + weekday", "m2: + events", "m3: + month + interactions"),
r2 = c(summary(m1)$r.squared, summary(m2)$r.squared, summary(m3)$r.squared),
adj_r2 = c(summary(m1)$adj.r.squared, summary(m2)$adj.r.squared, summary(m3)$adj.r.squared)
)

kable(model_compare, caption = "Model comparison (R² / Adjusted R²)")

Model comparison (R² / Adjusted R²)

model	r2	adj_r2
m1: trend + weekday	0.5456273	0.5364666
m2: + events	0.7335429	0.7225776
m3: + month + interactions	0.8312662	0.8127647

4.5 Model diagnostics

# # Quick diagnostic checks  see package is required

performance::check_model(m3)

5 Part 4 — Optional Direction : Lag Payday Effects

5.1 Create day-before and day-after payday variables

data <- data %>%
arrange(date) %>%
mutate(
payday_tomorrow = lead(payday, 1, default = 0),    # # 1 means: tomorrow is payday
payday_yesterday = lag(payday, 1, default = 0)     # # 1 means: yesterday was payday
)

m_lag <- lm(cust ~ time_index + daycat + payday + payday_tomorrow + payday_yesterday + holidays, data = data)
summary(m_lag)

## 
## Call:
## lm(formula = cust ~ time_index + daycat + payday + payday_tomorrow + 
##     payday_yesterday + holidays, data = data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -528.89 -103.69  -22.70   85.12  823.50 
## 
## Coefficients:
##                   Estimate Std. Error t value Pr(>|t|)    
## (Intercept)      930.01802   23.74766  39.163  < 2e-16 ***
## time_index         0.08548    0.15035   0.569   0.5702    
## daycat.L         163.54276   30.14359   5.425 1.39e-07 ***
## daycat.Q         323.63235   28.78388  11.244  < 2e-16 ***
## daycat.C          72.68755   28.66353   2.536   0.0118 *  
## daycat^4          48.85463   25.86246   1.889   0.0601 .  
## payday           468.13677   40.34813  11.602  < 2e-16 ***
## payday_tomorrow   32.36819   41.59557   0.778   0.4372    
## payday_yesterday 161.41237   39.85161   4.050 6.88e-05 ***
## holidays         304.67794   52.12808   5.845 1.62e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 173.8 on 244 degrees of freedom
## Multiple R-squared:  0.7497, Adjusted R-squared:  0.7405 
## F-statistic: 81.21 on 9 and 244 DF,  p-value: < 2.2e-16

kable(broom::tidy(m_lag), caption = "Lag model coefficients (tidy)")

Lag model coefficients (tidy)

term	estimate	std.error	statistic	p.value
(Intercept)	930.0180150	23.7476625	39.1625077	0.0000000
time_index	0.0854849	0.1503468	0.5685846	0.5701610
daycat.L	163.5427613	30.1435912	5.4254571	0.0000001
daycat.Q	323.6323492	28.7838825	11.2435266	0.0000000
daycat.C	72.6875457	28.6635301	2.5358895	0.0118412
daycat^4	48.8546260	25.8624555	1.8890173	0.0600754
payday	468.1367725	40.3481276	11.6024411	0.0000000
payday_tomorrow	32.3681899	41.5955650	0.7781644	0.4372260
payday_yesterday	161.4123691	39.8516054	4.0503354	0.0000688
holidays	304.6779437	52.1280786	5.8447952	0.0000000

6 Part 5 — AI-Powered Learning Journal

6.1 Prompt 1

“Explain how to interpret an interaction term in R regression (DAYCAT * Payday) in simple words.”

What I learned:

• Interaction means the payday effect changes depending on weekday.

• For example, a term like daycatFriday:payday means “extra payday impact on Fridays.”

6.2 Prompt 2

“Write ggplot code to plot daily visits over time, colored by Payday, and faceted by Holidays.”

What worked:

• Specifying my exact column names (date, cust, payday, holidays) helped.

How my prompt improved:

• I started including: dataset columns + the exact plot style I want (color, facet).

7 Conclusion

This study examined daily customer visits to a bank branch using event-based regression models that incorporate a time trend, weekday effects, and special events such as paydays and holidays.

Model 2 already explains a substantial portion of the variation in customer visits (Adjusted R² ≈ 0.723), indicating that weekday patterns and event flags provide strong predictive value. When monthly seasonality and interaction terms are added in Model 3, model fit improves further (Adjusted R² ≈ 0.813). This shows that incorporating seasonality significantly strengthens the model’s ability to explain daily visit behavior.

In Model 2, the time trend (time_index) is not statistically significant, suggesting there is no simple linear increase/decrease over time once weekday and event effects are controlled. However, after controlling for seasonality in Model 3, the time trend becomes negative and highly significant (time_index ≈ -12.0, p < 0.001), implying a modest underlying downward trend in visits within the seasonal pattern.

Weekday effects are strongly significant in both models, confirming that customer visits follow a systematic weekly rhythm rather than being evenly distributed across days. This indicates that operational planning should consider recurring weekday traffic differences.

Event variables provide clear managerial insights. Paydays have a large and statistically significant positive effect on visits in both models (about +443 in Model 2; about +376 in Model 3), meaning branches experience substantially higher traffic on payday periods. Holidays are also statistically significant (about +302 in Model 2; about +298 in Model 3), showing that customer visits differ meaningfully on holidays compared to normal days in the dataset.

In contrast, staff payday (sp), faculty payday (fac), and before-holiday (bh) are not statistically significant, suggesting these factors do not materially affect total visits after controlling for the main payday/holiday effects. Additionally, the interaction terms (weekday × payday and holiday × faculty payday) are not statistically significant, implying the payday effect is relatively consistent across weekdays rather than being concentrated on a specific day of the week.

Finally, monthly seasonality is one of the strongest drivers in Model 3, with month effects increasing substantially from February through December. This indicates pronounced seasonal variation in branch traffic and suggests that staffing and service capacity planning should consider both payday peaks and high-traffic months.

Overall, the results demonstrate that an event-based modeling approach combined with seasonality provides a strong framework for understanding and predicting bank customer visits, supporting better operational and staffing decisions.