Identifying frequently purchased groceries
Mohamed Ali Hefnawy
8/7/2020
Introduction
This project is from Book: Machine learning with R by Brett Lantz, chapter 8.
A link to the book https://bit.ly/3gsf2e0
This project is for educational purpose only.
The aim is to perform a market basket analysis of transnational data from a grocery store.
Required packages
we will use arules package
library(arules)## Loading required package: Matrix##
## Attaching package: 'arules'## The following objects are masked from 'package:base':
##
## abbreviate, writeSetp 1 - collecting data
The dataset was adapted from the Groceries dataset in arules R package.
Step 2 - exploring and preparing the data
As the data of each purchase are separated by a comma, it is not feasible to read the file using read.csv, instead we will use arules package to create a spares matrix for transaction data.
Data preparation - creating a spares matrix for transaction data
groceries <- read.transactions("groceries.csv", sep = ",")
#Exploring the summary of the spares matrix
summary(groceries)## transactions as itemMatrix in sparse format with
## 9835 rows (elements/itemsets/transactions) and
## 169 columns (items) and a density of 0.02609146
##
## most frequent items:
## whole milk other vegetables rolls/buns soda
## 2513 1903 1809 1715
## yogurt (Other)
## 1372 34055
##
## element (itemset/transaction) length distribution:
## sizes
## 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
## 2159 1643 1299 1005 855 645 545 438 350 246 182 117 78 77 55 46
## 17 18 19 20 21 22 23 24 26 27 28 29 32
## 29 14 14 9 11 4 6 1 1 1 1 3 1
##
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.000 2.000 3.000 4.409 6.000 32.000
##
## includes extended item information - examples:
## labels
## 1 abrasive cleaner
## 2 artif. sweetener
## 3 baby cosmeticsWe can see dimension of the spares matrix, 169 columns X 9835 rows. 169 columns represents 169 different items that might appear in someone’s grocery basket. Each cell in the matrix is a 1 if the item was purchased for the corresponding transaction, or 0 otherwise.
The Density value 0.026 represents the proportion of non-zero matrix cells. Below it we can see frequently purchased items. finally we can find some statistics about the size of the transactions. A total of 2159 transactions have 1 item.
#we can inspect more the data using inspect and item frequency
inspect(groceries[1:5])## items
## [1] {citrus fruit,
## margarine,
## ready soups,
## semi-finished bread}
## [2] {coffee,
## tropical fruit,
## yogurt}
## [3] {whole milk}
## [4] {cream cheese,
## meat spreads,
## pip fruit,
## yogurt}
## [5] {condensed milk,
## long life bakery product,
## other vegetables,
## whole milk}itemFrequency(groceries[, 1:3])## abrasive cleaner artif. sweetener baby cosmetics
## 0.0035587189 0.0032536858 0.0006100661We notice that items are stored by alphabetical order, the first one is found in about 0.3 percent of the transactions
Visualizing item support - item frequency plot
itemFrequencyPlot(groceries, support = 0.1)
To know the top items
itemFrequencyPlot(groceries, topN = 20)
We can notice that milk and other vegetables are on the top of the list.
Visualizing the transaction data - plotting the spares matrix
we will use image to have a bird’s-eye view of the entire sparse matrix.
#below code for the first 5 transactions.
image(groceries[1:5])
To see random sample of 100 transactions
image(sample(groceries, 100))
We can see the common items between the transactions.
Step 3 - training a model on the data
we will use apriori function to train the model, but we have to select parameters like support, confidence and minlen in a proper way to help us filter the required data only.
groceryrules <- apriori(groceries, parameter = list(support = 0.006, confidence = 0.25, minlen = 2))## Apriori
##
## Parameter specification:
## confidence minval smax arem aval originalSupport maxtime support minlen
## 0.25 0.1 1 none FALSE TRUE 5 0.006 2
## maxlen target ext
## 10 rules TRUE
##
## Algorithmic control:
## filter tree heap memopt load sort verbose
## 0.1 TRUE TRUE FALSE TRUE 2 TRUE
##
## Absolute minimum support count: 59
##
## set item appearances ...[0 item(s)] done [0.00s].
## set transactions ...[169 item(s), 9835 transaction(s)] done [0.00s].
## sorting and recoding items ... [109 item(s)] done [0.00s].
## creating transaction tree ... done [0.00s].
## checking subsets of size 1 2 3 4 done [0.00s].
## writing ... [463 rule(s)] done [0.00s].
## creating S4 object ... done [0.00s].groceryrules## set of 463 rulesWe have 463 rules now, we need to know which of them is useful to us.
Step 4 - evaluating model performance
summary(groceryrules)## set of 463 rules
##
## rule length distribution (lhs + rhs):sizes
## 2 3 4
## 150 297 16
##
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 2.000 2.000 3.000 2.711 3.000 4.000
##
## summary of quality measures:
## support confidence coverage lift
## Min. :0.006101 Min. :0.2500 Min. :0.009964 Min. :0.9932
## 1st Qu.:0.007117 1st Qu.:0.2971 1st Qu.:0.018709 1st Qu.:1.6229
## Median :0.008744 Median :0.3554 Median :0.024809 Median :1.9332
## Mean :0.011539 Mean :0.3786 Mean :0.032608 Mean :2.0351
## 3rd Qu.:0.012303 3rd Qu.:0.4495 3rd Qu.:0.035892 3rd Qu.:2.3565
## Max. :0.074835 Max. :0.6600 Max. :0.255516 Max. :3.9565
## count
## Min. : 60.0
## 1st Qu.: 70.0
## Median : 86.0
## Mean :113.5
## 3rd Qu.:121.0
## Max. :736.0
##
## mining info:
## data ntransactions support confidence
## groceries 9835 0.006 0.25To take closer look on specific rules
inspect(groceryrules[1:3])## lhs rhs support confidence coverage
## [1] {potted plants} => {whole milk} 0.006914082 0.4000000 0.01728521
## [2] {pasta} => {whole milk} 0.006100661 0.4054054 0.01504830
## [3] {herbs} => {root vegetables} 0.007015760 0.4312500 0.01626843
## lift count
## [1] 1.565460 68
## [2] 1.586614 60
## [3] 3.956477 69We want to divide the rules to three categories, actionable, trivial, and inexplicable.
Step 5 - improving model performance
Sorting the set of association rules
We will sort the rules by lift
inspect(sort(groceryrules, by = "lift") [1:5])## lhs rhs support confidence coverage lift count
## [1] {herbs} => {root vegetables} 0.007015760 0.4312500 0.01626843 3.956477 69
## [2] {berries} => {whipped/sour cream} 0.009049314 0.2721713 0.03324860 3.796886 89
## [3] {other vegetables,
## tropical fruit,
## whole milk} => {root vegetables} 0.007015760 0.4107143 0.01708185 3.768074 69
## [4] {beef,
## other vegetables} => {root vegetables} 0.007930859 0.4020619 0.01972547 3.688692 78
## [5] {other vegetables,
## tropical fruit} => {pip fruit} 0.009456024 0.2634561 0.03589222 3.482649 93from first rule we can say that people who buy herbs are nearly four times more likely to buy root vegetables than the typical customer.
Taking subsets of association rules
berryrules <- subset(groceryrules, items %in% "berries")
inspect(berryrules)## lhs rhs support confidence coverage lift
## [1] {berries} => {whipped/sour cream} 0.009049314 0.2721713 0.0332486 3.796886
## [2] {berries} => {yogurt} 0.010574479 0.3180428 0.0332486 2.279848
## [3] {berries} => {other vegetables} 0.010269446 0.3088685 0.0332486 1.596280
## [4] {berries} => {whole milk} 0.011794611 0.3547401 0.0332486 1.388328
## count
## [1] 89
## [2] 104
## [3] 101
## [4] 116Saving association rules to a file or a data frame
write(groceryrules, file = "groceryrules.csv", sep = ",", quote = TRUE, row.names = FALSE)
groceryrules_df <- as(groceryrules, "data.frame")
str(groceryrules_df)## 'data.frame': 463 obs. of 6 variables:
## $ rules : chr "{potted plants} => {whole milk}" "{pasta} => {whole milk}" "{herbs} => {root vegetables}" "{herbs} => {other vegetables}" ...
## $ support : num 0.00691 0.0061 0.00702 0.00773 0.00773 ...
## $ confidence: num 0.4 0.405 0.431 0.475 0.475 ...
## $ coverage : num 0.0173 0.015 0.0163 0.0163 0.0163 ...
## $ lift : num 1.57 1.59 3.96 2.45 1.86 ...
## $ count : int 68 60 69 76 76 69 70 67 63 88 ...Summary
Association rules helps to find out which items can be purchased in association with other items, this may help wholesalers to find best combinations for promotions.