Market Basket Analysis

Ever wondered why stores like Easyday,Big Bazar place confectioneries near the checkout counter ? Or why the dairy products section is always near one corner of the store ? While some answers may seem highly intuitive, the fact of the matter is most are not !

These stores make use of analytics by analyzing your transactions , in order to guess which items go together in the shopping basket.Retailers can use this data in countless ways to increase their revenue.

For example -

1. Store Layout planning (Placing products that go together close by to increase spend per shopping basket and to increase customer shopping experience)

2. Targeted Marketing (Using the data about customers collected to inform them of latest offering which might be of their interest)

This post will try to explain the basic concept behind this form of retail analytics. I have performed the market basket analysis using the apriori algorithm in R.

The dataset for this can be downloaded from here: https://www.dropbox.com/s/n09hu3mr63pq4bw/skusold.csv

An intro into Apriori Algorithm:

Courtesy: http://webdocs.cs.ualberta.ca/~zaiane/courses/cmput499/slides/Lect10/img054.jpg

Follow the comments alongside the code to understand the flow of logic.

*Text in white is the code.

*Text in blue is the output.

#Market Basket Analysis 

#Installing package  'arules' for apriori algorithm
install.packages("arules")
library(arules)install.packages("knitr")
library(knitr)
#Checking working directory
getwd()
[1] "C:/Users/Shanky/Documents"
#Reading the file
tt<-read.csv("skusold.csv")
head(tt)

#It can be seen that there are multiple orders from the same Order ID. 
Thus, we need to now arrange them in such a manner that all orders pertaining to one transaction are listed together.
  Order       SKU
1  9305     Bread
2  9305 Tropicana
3  9305      Eggs
4 11020 Tropicana
5 11020  Gilette 
6 11020      Eggs

#Our objective is to group or aggregate the items together based on the Order ID. We can do that as follows:
Aggdata<-split(tt$SKU,tt$Order)
head(Aggdata)

#All items bought from same order ID listed
$`9305`
[1] Bread     Tropicana Eggs     
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana

$`11013`
[1] Bread     Tropicana Jam      
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana

$`11015`
[1] Eggs  Bread
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana

$`11017`
[1] Tomatoes   AfterShave Tropicana  Gilette    Eggs      
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana

$`11018`
[1] Jam        AfterShave Tropicana  Tomatoes  
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana

$`11019`
[1] AfterShave Jam        Tomatoes   Tropicana  Hide       Gilette   
Levels: AfterShave Bread Eggs Gilette  Hide Jam Tomatoes Tropicana
#For using the Apriori algorithm, we need to coerce the transaction
abc = as(Aggdata,"transactions")
#Checking the summary statistics of the data
summary(abc)
transactions as itemMatrix in sparse format with
 13 rows (elements/itemsets/transactions) and
 8 columns (items) and a density of 0.4712 

most frequent items:
     Bread  Tropicana AfterShave       Eggs   Gilette     (Other) 
         8          8          7          7          7         12 

element (itemset/transaction) length distribution:
sizes
2 3 4 5 6 
2 4 3 3 1 

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
   2.00    3.00    4.00    3.77    5.00    6.00 

includes extended item information - examples:
      labels
1 AfterShave
2      Bread
3       Eggs
includes extended transaction information - examples:
  transactionID
1          9305
2         11013
3         11015

#The transaction an itemMatrix forms the input for the Apriori Algorithm
Rules<-apriori(abc,parameter=list(supp=0.3,conf=0.8,target="rules",minlen=2))

parameter specification:
 confidence minval smax arem  aval originalSupport support minlen maxlen
        0.8    0.1    1 none FALSE            TRUE     0.3      2     10
 target   ext
  rules FALSE

algorithmic control:
 filter tree heap memopt load sort verbose
    0.1 TRUE TRUE  FALSE TRUE    2    TRUE

apriori - find association rules with the apriori algorithm

#To view the Rules generated
inspect(Rules)
  lhs            rhs          support confidence  lift
1 {Tomatoes}  => {AfterShave}  0.3077     1.0000 1.857
2 {Jam}       => {Tropicana}   0.3846     0.8333 1.354
3 {Eggs}      => {Bread}       0.4615     0.8571 1.393
4 {Eggs,                                              
   Tropicana} => {Bread}       0.3077     0.8000 1.300
5 {Bread,                                             
   Tropicana} => {Eggs}        0.3077     0.8000 1.486
#To inspect items by frequency
itemFrequencyPlot(abc,topN=5,col="red")

Interpretation of the Rules
lhs                  rhs       support    confidence     lift

{Eggs} => {Bread}   0.4615      0.857             1 1.393

The support indicates the percentage of occurrence of this transaction in all possible transactions.
The confidence indicates the correctness of prediction and likelihood of purchase.
In this example, if a customer buys eggs then the likelihood that he will buy bread as well is ~ 85%. The support indicates that ~46% of transactions had similar characteristics.

References
1.Association Rule Mining using R package
2.Wikipedia - Apriori Algorithm3.Maching learning