Unsupervised Learning Project: Association Rules
Aleksandra Dobosz
2025-01-26
Association Rules - Census Income Dataset
Dataset overview
The Adult Dataset from the UCI Machine Learning Repository. It was
extracted from the 1994 U.S. Census database and is also known as the
“Census Income Dataset.” The dataset contains both categorical and
numerical variables:
1. Age (numerical) – Age of the individual.
2. Workclass (categorical) – Type of employer (e.g., private,
self-employed, government).
3. Fnlwgt (numerical) – Final weight (adjusted for population
sampling).
4. Education (categorical) – Highest level of education attained.
5. Education-num (numerical) – Education level encoded as a numeric
variable.
6. Marital-status (categorical) – Marital status (e.g., married,
divorced, never married).
7. Occupation (categorical) – Job category (e.g., tech support, sales,
clerical).
8. Relationship (categorical) – Relationship to household head (e.g.,
husband, wife, unmarried).
9. Race (categorical) – Race of the individual.
10. Sex (categorical) – Gender (Male/Female).
11. Capital-gain (numerical) – Income from capital gains.
12. Capital-loss (numerical) – Income lost due to capital losses.
13. Hours-per-week (numerical) – Weekly working hours.
14. Native-country (categorical) – Country of origin.
15. Income (categorical) – Either “<=50K” or “>50K” (annual income
in $).
Goal of this paper
The primary objective of this paper is to apply association rules in order to find hidden relationships between demographic, occupational, and financial attributes. Association rules help identify patterns such as: Which education levels are strongly associated with high income? What work classes frequently appear with specific occupations? Are certain marital statuses more likely to be linked with a higher number of weekly work hours? I will pay special attention to find patterns that strongly indicate high-income and low income individuals. This will be achieved through:
- Apriori Algorithm, which is a foundational method for discovering frequent items and generating association rules. It is characterized with its significant ability to identify this relationships in large datasets.1
Preprocessing the data
Firstly, I am preprocessing the data - changing column names, removing NA’s, converting categorical variables into factors.
col_names <- c('Age', 'WorkClass', 'fnlwgt', 'Education', 'EducationNum', 'MaritalStatus', 'Occupation',
'Relationship', 'Race', 'Gender', 'CapitalGain', 'CapitalLoss', 'HoursPerWeek', 'NativeCountry', 'Income')
data <- read.csv("adult.csv", na.strings = "?")
head(data, 10)## age workclass fnlwgt education educational.num marital.status
## 1 25 Private 226802 11th 7 Never-married
## 2 38 Private 89814 HS-grad 9 Married-civ-spouse
## 3 28 Local-gov 336951 Assoc-acdm 12 Married-civ-spouse
## 4 44 Private 160323 Some-college 10 Married-civ-spouse
## 5 18 <NA> 103497 Some-college 10 Never-married
## 6 34 Private 198693 10th 6 Never-married
## 7 29 <NA> 227026 HS-grad 9 Never-married
## 8 63 Self-emp-not-inc 104626 Prof-school 15 Married-civ-spouse
## 9 24 Private 369667 Some-college 10 Never-married
## 10 55 Private 104996 7th-8th 4 Married-civ-spouse
## occupation relationship race gender capital.gain capital.loss
## 1 Machine-op-inspct Own-child Black Male 0 0
## 2 Farming-fishing Husband White Male 0 0
## 3 Protective-serv Husband White Male 0 0
## 4 Machine-op-inspct Husband Black Male 7688 0
## 5 <NA> Own-child White Female 0 0
## 6 Other-service Not-in-family White Male 0 0
## 7 <NA> Unmarried Black Male 0 0
## 8 Prof-specialty Husband White Male 3103 0
## 9 Other-service Unmarried White Female 0 0
## 10 Craft-repair Husband White Male 0 0
## hours.per.week native.country income
## 1 40 United-States <=50K
## 2 50 United-States <=50K
## 3 40 United-States >50K
## 4 40 United-States >50K
## 5 30 United-States <=50K
## 6 30 United-States <=50K
## 7 40 United-States <=50K
## 8 32 United-States >50K
## 9 40 United-States <=50K
## 10 10 United-States <=50Kcolnames(data) <- col_names
data <- na.omit(data)
#WorkClass, Education, MaritalStatus, Occupation, Relationship, Race, Gender, NativeCountry, Income are characters
characters <- c('WorkClass', 'MaritalStatus', 'Occupation', 'Relationship', 'Race',
'Gender', 'NativeCountry', 'Income')
data[characters] <- lapply(data[characters], as.factor)
data$Education <- cut(data$EducationNum, breaks = c(0, 5, 10, 15, 16), labels = c("Low", "Medium", "High", "Very High"), right = TRUE)
data$Age <- cut(data$Age, breaks = c(17, 30, 40, 50, 65, 90), labels = c("17-30", "31-40", "41-50", "51-65", "66-90"), right = TRUE)
data$CapitalGain <- cut(data$CapitalGain, breaks = c(-Inf, 1000, 5000, 20000, 50000, Inf), labels = c("Very Low", "Low", "Medium", "High", "Very High"))
data$CapitalLoss <- cut(data$CapitalLoss, breaks = 5, labels = c("Very Low", "Low", "Medium", "High", "Very High"), include.lowest = TRUE)
data$HoursPerWeek <- cut(data$HoursPerWeek, breaks = c(0, 20, 40, 60, 80, 100), labels = c("Very Low", "Low", "Medium", "High", "Very High"), right = TRUE)
data <- subset(data, select = -c(EducationNum, fnlwgt))
str(data)## 'data.frame': 45222 obs. of 13 variables:
## $ Age : Factor w/ 5 levels "17-30","31-40",..: 1 2 1 3 2 4 1 4 4 2 ...
## $ WorkClass : Factor w/ 7 levels "Federal-gov",..: 3 3 2 3 3 5 3 3 3 1 ...
## $ Education : Factor w/ 4 levels "Low","Medium",..: 2 2 3 2 2 3 2 1 2 3 ...
## $ MaritalStatus: Factor w/ 7 levels "Divorced","Married-AF-spouse",..: 5 3 3 3 5 3 5 3 3 3 ...
## $ Occupation : Factor w/ 14 levels "Adm-clerical",..: 7 5 11 7 8 10 8 3 7 1 ...
## $ Relationship : Factor w/ 6 levels "Husband","Not-in-family",..: 4 1 1 1 2 1 5 1 1 1 ...
## $ Race : Factor w/ 5 levels "Amer-Indian-Eskimo",..: 3 5 5 3 5 5 5 5 5 5 ...
## $ Gender : Factor w/ 2 levels "Female","Male": 2 2 2 2 2 2 1 2 2 2 ...
## $ CapitalGain : Factor w/ 5 levels "Very Low","Low",..: 1 1 1 3 1 2 1 1 3 1 ...
## $ CapitalLoss : Factor w/ 5 levels "Very Low","Low",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ HoursPerWeek : Factor w/ 5 levels "Very Low","Low",..: 2 3 2 2 2 2 2 1 2 2 ...
## $ NativeCountry: Factor w/ 41 levels "Cambodia","Canada",..: 39 39 39 39 39 39 39 39 39 39 ...
## $ Income : Factor w/ 2 levels "<=50K",">50K": 1 1 2 2 1 2 1 1 2 1 ...Now, I am transforming my data into format which is suitable for Apriori algorithm.
library(arules)
transactions_income <- as(data, "transactions")summary(transactions_income)## transactions as itemMatrix in sparse format with
## 45222 rows (elements/itemsets/transactions) and
## 108 columns (items) and a density of 0.1202694
##
## most frequent items:
## CapitalLoss=Very Low CapitalGain=Very Low
## 43117 41498
## NativeCountry=United-States Race=White
## 41292 38903
## Income=<=50K (Other)
## 34014 388569
##
## element (itemset/transaction) length distribution:
## sizes
## 12 13
## 493 44729
##
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 12.00 13.00 13.00 12.99 13.00 13.00
##
## includes extended item information - examples:
## labels variables levels
## 1 Age=17-30 Age 17-30
## 2 Age=31-40 Age 31-40
## 3 Age=41-50 Age 41-50
##
## includes extended transaction information - examples:
## transactionID
## 1 1
## 2 2
## 3 3inspect(transactions_income[1:10])## items transactionID
## [1] {Age=17-30,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Never-married,
## Occupation=Machine-op-inspct,
## Relationship=Own-child,
## Race=Black,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=<=50K} 1
## [2] {Age=31-40,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Married-civ-spouse,
## Occupation=Farming-fishing,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Medium,
## NativeCountry=United-States,
## Income=<=50K} 2
## [3] {Age=17-30,
## WorkClass=Local-gov,
## Education=High,
## MaritalStatus=Married-civ-spouse,
## Occupation=Protective-serv,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=>50K} 3
## [4] {Age=41-50,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Married-civ-spouse,
## Occupation=Machine-op-inspct,
## Relationship=Husband,
## Race=Black,
## Gender=Male,
## CapitalGain=Medium,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=>50K} 4
## [5] {Age=31-40,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Never-married,
## Occupation=Other-service,
## Relationship=Not-in-family,
## Race=White,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=<=50K} 6
## [6] {Age=51-65,
## WorkClass=Self-emp-not-inc,
## Education=High,
## MaritalStatus=Married-civ-spouse,
## Occupation=Prof-specialty,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=>50K} 8
## [7] {Age=17-30,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Never-married,
## Occupation=Other-service,
## Relationship=Unmarried,
## Race=White,
## Gender=Female,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=<=50K} 9
## [8] {Age=51-65,
## WorkClass=Private,
## Education=Low,
## MaritalStatus=Married-civ-spouse,
## Occupation=Craft-repair,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Very Low,
## NativeCountry=United-States,
## Income=<=50K} 10
## [9] {Age=51-65,
## WorkClass=Private,
## Education=Medium,
## MaritalStatus=Married-civ-spouse,
## Occupation=Machine-op-inspct,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Medium,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=>50K} 11
## [10] {Age=31-40,
## WorkClass=Federal-gov,
## Education=High,
## MaritalStatus=Married-civ-spouse,
## Occupation=Adm-clerical,
## Relationship=Husband,
## Race=White,
## Gender=Male,
## CapitalGain=Very Low,
## CapitalLoss=Very Low,
## HoursPerWeek=Low,
## NativeCountry=United-States,
## Income=<=50K} 12itemFrequencyPlot(transactions_income, support = 0.1, col="coral4")
Apriori Algorithm
I am now applying Apriori Algorithm.I am setting support level 10% - this means that an item must appear in at least 10% of the dataset. It will help me filter out rare, less meaningful patterns. Confidence is set at level of 80% - whenever the antecedent (left-hand side) occur, there is at least 80% chances that the consequent (right-hand side) will occur. After applying the Apriori algorithm, I am sorting by lift, confidence and count. Sorting by lift find strong dependencies between attributes beyond random occurence. It can be used in relation to high income. Confidence sorting displays how reliable the rule is in predicting an outcome. Finally, support sorting is used to find frquent patterns that appear in many transactions.
rules <- apriori(transactions_income, parameter = list(support = 0.1, confidence = 0.8), maxlen = 3)## Apriori
##
## Parameter specification:
## confidence minval smax arem aval originalSupport maxtime support minlen
## 0.8 0.1 1 none FALSE TRUE 5 0.1 1
## maxlen target ext
## 3 rules TRUE
##
## Algorithmic control:
## filter tree heap memopt load sort verbose
## 0.1 TRUE TRUE FALSE TRUE 2 TRUE
##
## Absolute minimum support count: 4522
##
## set item appearances ...[0 item(s)] done [0.00s].
## set transactions ...[108 item(s), 45222 transaction(s)] done [0.02s].
## sorting and recoding items ... [30 item(s)] done [0.00s].
## creating transaction tree ... done [0.01s].
## checking subsets of size 1 2 3## done [0.01s].
## writing ... [946 rule(s)] done [0.00s].
## creating S4 object ... done [0.00s].summary(rules)## set of 946 rules
##
## rule length distribution (lhs + rhs):sizes
## 1 2 3
## 4 133 809
##
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.000 3.000 3.000 2.851 3.000 3.000
##
## summary of quality measures:
## support confidence coverage lift
## Min. :0.1002 Min. :0.8002 Min. :0.1029 Min. :0.8725
## 1st Qu.:0.1308 1st Qu.:0.8847 1st Qu.:0.1440 1st Qu.:0.9934
## Median :0.2046 Median :0.9177 Median :0.2238 Median :1.0172
## Mean :0.2587 Mean :0.9144 Mean :0.2837 Mean :1.1066
## 3rd Qu.:0.2991 3rd Qu.:0.9477 3rd Qu.:0.3228 3rd Qu.:1.0591
## Max. :0.9535 Max. :1.0000 Max. :1.0000 Max. :2.9114
## count
## Min. : 4531
## 1st Qu.: 5913
## Median : 9253
## Mean :11700
## 3rd Qu.:13527
## Max. :43117
##
## mining info:
## data ntransactions support confidence
## transactions_income 45222 0.1 0.8
## call
## apriori(data = transactions_income, parameter = list(support = 0.1, confidence = 0.8), maxlen = 3)inspect(sort(rules, by = "lift")[1:10])## lhs rhs support confidence coverage lift count
## [1] {Age=17-30,
## Relationship=Own-child} => {MaritalStatus=Never-married} 0.1022290 0.9398252 0.1087745 2.911411 4623
## [2] {WorkClass=Private,
## Relationship=Own-child} => {MaritalStatus=Never-married} 0.1107868 0.8927299 0.1240989 2.765518 5010
## [3] {Relationship=Own-child,
## Race=White} => {MaritalStatus=Never-married} 0.1100128 0.8909384 0.1234797 2.759968 4975
## [4] {Relationship=Own-child,
## Income=<=50K} => {MaritalStatus=Never-married} 0.1282783 0.8895875 0.1441997 2.755783 5801
## [5] {Relationship=Own-child,
## NativeCountry=United-States} => {MaritalStatus=Never-married} 0.1221529 0.8876748 0.1376100 2.749858 5524
## [6] {Relationship=Own-child,
## CapitalGain=Very Low} => {MaritalStatus=Never-married} 0.1266640 0.8875116 0.1427181 2.749353 5728
## [7] {Relationship=Own-child,
## CapitalLoss=Very Low} => {MaritalStatus=Never-married} 0.1269294 0.8860759 0.1432489 2.744905 5740
## [8] {Relationship=Own-child} => {MaritalStatus=Never-married} 0.1296714 0.8849985 0.1465216 2.741567 5864
## [9] {Education=Medium,
## Relationship=Own-child} => {MaritalStatus=Never-married} 0.1022069 0.8845933 0.1155411 2.740312 4622
## [10] {MaritalStatus=Married-civ-spouse,
## Gender=Male} => {Relationship=Husband} 0.4124983 0.9900223 0.4166556 2.398521 18654inspect(sort(rules, by = "confidence")[1:10])## lhs rhs support confidence coverage lift count
## [1] {Age=41-50,
## Relationship=Husband} => {Gender=Male} 0.1194109 1.0000000 0.1194109 1.481377 5400
## [2] {Relationship=Husband,
## Income=>50K} => {Gender=Male} 0.1881164 1.0000000 0.1881164 1.481377 8507
## [3] {Relationship=Husband,
## HoursPerWeek=Medium} => {Gender=Male} 0.1543054 1.0000000 0.1543054 1.481377 6978
## [4] {Education=High,
## Relationship=Husband} => {Gender=Male} 0.1471629 1.0000000 0.1471629 1.481377 6655
## [5] {Relationship=Husband} => {Gender=Male} 0.4127416 0.9999464 0.4127637 1.481298 18665
## [6] {MaritalStatus=Married-civ-spouse,
## Relationship=Husband} => {Gender=Male} 0.4124983 0.9999464 0.4125205 1.481298 18654
## [7] {Relationship=Husband,
## CapitalLoss=Very Low} => {Gender=Male} 0.3861616 0.9999427 0.3861837 1.481292 17463
## [8] {Relationship=Husband,
## NativeCountry=United-States} => {Gender=Male} 0.3776702 0.9999415 0.3776923 1.481290 17079
## [9] {Relationship=Husband,
## Race=White} => {Gender=Male} 0.3751935 0.9999411 0.3752156 1.481290 16967
## [10] {Relationship=Husband,
## CapitalGain=Very Low} => {Gender=Male} 0.3631639 0.9999391 0.3631861 1.481287 16423inspect(sort(rules, by = "count")[1:10])## lhs rhs support
## [1] {} => {CapitalLoss=Very Low} 0.9534519
## [2] {} => {CapitalGain=Very Low} 0.9176507
## [3] {} => {NativeCountry=United-States} 0.9130954
## [4] {CapitalGain=Very Low} => {CapitalLoss=Very Low} 0.8711026
## [5] {CapitalLoss=Very Low} => {CapitalGain=Very Low} 0.8711026
## [6] {NativeCountry=United-States} => {CapitalLoss=Very Low} 0.8698421
## [7] {CapitalLoss=Very Low} => {NativeCountry=United-States} 0.8698421
## [8] {} => {Race=White} 0.8602671
## [9] {NativeCountry=United-States} => {CapitalGain=Very Low} 0.8364955
## [10] {CapitalGain=Very Low} => {NativeCountry=United-States} 0.8364955
## confidence coverage lift count
## [1] 0.9534519 1.0000000 1.0000000 43117
## [2] 0.9176507 1.0000000 1.0000000 41498
## [3] 0.9130954 1.0000000 1.0000000 41292
## [4] 0.9492747 0.9176507 0.9956189 39393
## [5] 0.9136304 0.9534519 0.9956189 39393
## [6] 0.9526300 0.9130954 0.9991381 39336
## [7] 0.9123084 0.9534519 0.9991381 39336
## [8] 0.8602671 1.0000000 1.0000000 38903
## [9] 0.9161097 0.9130954 0.9983207 37828
## [10] 0.9115620 0.9176507 0.9983207 37828What drives low and high income?
Now, my Apriori implementation is generating association rules where the right-hand side is fixed to “Income=<=50K”, meaning you’re finding patterns that strongly indicate low-income individuals. I am sorting by “lift”, because I want meanignful, strong assosiations. I am setting the support level at 0.01 to catch rare but more meaningful patterns, and confidence at 0.5 to reduce redundant rules.
#What drives the low income
rules.less50K<-apriori(data=transactions_income, parameter=list(supp=0.01,conf = 0.5),
appearance=list(default="lhs", rhs="Income=<=50K"), control=list(verbose=F))
rules.less50K.bylift<-sort(rules.less50K, by="lift", decreasing=TRUE)
inspect(head(rules.less50K.bylift))## lhs rhs support confidence coverage lift count
## [1] {Occupation=Other-service,
## Relationship=Own-child,
## HoursPerWeek=Very Low} => {Income=<=50K} 0.01090177 1 0.01090177 1.329511 493
## [2] {MaritalStatus=Never-married,
## Occupation=Other-service,
## HoursPerWeek=Very Low} => {Income=<=50K} 0.01364380 1 0.01364380 1.329511 617
## [3] {Relationship=Own-child,
## Gender=Male,
## HoursPerWeek=Very Low} => {Income=<=50K} 0.01581089 1 0.01581089 1.329511 715
## [4] {Relationship=Own-child,
## CapitalGain=Very Low,
## HoursPerWeek=Very Low} => {Income=<=50K} 0.03498297 1 0.03498297 1.329511 1582
## [5] {Age=17-30,
## Occupation=Handlers-cleaners,
## Relationship=Own-child,
## CapitalGain=Very Low} => {Income=<=50K} 0.01068064 1 0.01068064 1.329511 483
## [6] {MaritalStatus=Never-married,
## Occupation=Handlers-cleaners,
## Relationship=Own-child,
## Gender=Male} => {Income=<=50K} 0.01112290 1 0.01112290 1.329511 503Analyzing the rules I can say that there are some variables that seem to be highly associated with earning less than or equal $50K in this dataset. Occupation (such as “Other-service” and “Handlers-cleaners”), hours worked (“Very Low”), relationship status (“Own-child”, “Never-married”), capital gains (“Very Low”) and age (17-30 years). In this case, all rules have 100% confidence, meaning that when the conditions on the LHS hold, the RHS (Income <= 50K) will always hold. This suggests that these combinations are strong predictors of income being less than or equal to 50K. The lift value for all rules is the same (1.329511), indicating that these rules are 32.95% more likely to occur than by chance.
Below, I am generating association rules where the right-hand side is fixed to “Income=>50K”, meaning finding patterns that strongly indicate high-income individuals.
#What drives high income
rules.more50K<-apriori(data=transactions_income, parameter=list(supp=0.01,conf = 0.5),
appearance=list(default="lhs", rhs="Income=>50K"), control=list(verbose=F))
rules.more50K.bylift<-sort(rules.more50K, by="lift", decreasing=TRUE)
inspect(head(rules.more50K.bylift))## lhs rhs support confidence coverage lift count
## [1] {Occupation=Exec-managerial,
## Race=White,
## CapitalGain=Medium} => {Income=>50K} 0.01023838 0.9585921 0.01068064 3.867724 463
## [2] {Occupation=Exec-managerial,
## Race=White,
## CapitalGain=Medium,
## CapitalLoss=Very Low} => {Income=>50K} 0.01023838 0.9585921 0.01068064 3.867724 463
## [3] {Education=High,
## MaritalStatus=Married-civ-spouse,
## CapitalGain=Medium,
## NativeCountry=United-States} => {Income=>50K} 0.01742515 0.9563107 0.01822122 3.858519 788
## [4] {Education=High,
## MaritalStatus=Married-civ-spouse,
## CapitalGain=Medium,
## CapitalLoss=Very Low,
## NativeCountry=United-States} => {Income=>50K} 0.01742515 0.9563107 0.01822122 3.858519 788
## [5] {Education=High,
## MaritalStatus=Married-civ-spouse,
## CapitalGain=Medium} => {Income=>50K} 0.01864137 0.9557823 0.01950378 3.856387 843
## [6] {Education=High,
## MaritalStatus=Married-civ-spouse,
## CapitalGain=Medium,
## CapitalLoss=Very Low} => {Income=>50K} 0.01864137 0.9557823 0.01950378 3.856387 843These rules are explaining associations between various factors and the likelihood of earning greater than $50K.
The strong inidicators of high income revealed from the rules above are:
Occupation: “Exec-managerial” jobs are highly
associated with earning more than $50K.
Race: Being “White” is seen frequently in these
high-income rules.
Capital Gain: Individuals with “Medium” capital gain
are often associated with higher incomes.
Education: Having “High” education (such as a high
school diploma, college, or advanced degrees) is an important factor for
earning more than $50K.
Marital Status: Being “Married-civ-spouse” also
correlates strongly with earning more than $50K.
Native Country: Being from the United States also
increases the likelihood of earning more than $50K.
The lift values for these rules are consistently high (around 3.86), meaning that these rules are significantly more likely than random chance to occur in the dataset. This suggests that the combinations of these attributes (e.g., occupation, education, capital gain, marital status) are very strong predictors of income being above $50K.
library(arulesViz)
itemFrequencyPlot(transactions_income, topN=10, type="absolute", main="Item Frequency - absolute", col="coral") 
itemFrequencyPlot(transactions_income, topN=10, type="relative", main="Item Frequency - relative", col="coral4")
plot(rules, method = "grouped")
First plot above shows the top 10 most frequent items in my dataset. The second plot displays the proportion of occureness of each item relative to all of them. the third one, visualizes the association rules generated using Apriori algorithm. Shows relationship between left-hand side and right-hand side of the rules. Now I will display an interactive network graph of rules.
# income <=50K
plot(rules.less50K, method="graph", control = list(max = 10))
# income >50K
plot(rules.more50K, method="graph", control = list(max = 10))
Conclusions
By mining association rules, I identified key factors that strongly influence income levels - distinguishing between individuals earning above or below $50K per year. This analysis indicates that low-income individuals are commonly associated with low education levels, service-sector occupations, young age, and low capital gains, whereas high-income individuals tend to have managerial or executive roles, high education levels, significant capital gains, and are often married. The strong lift values of these rules confirm that these relationships are far from random and can provide meaningful insights.