Association Rule Mining : Example & R code

Let's try to solve the following association rules mining problem.

Minimum support = 30%

Solution:  Association Rules Mining

Transaction ID	Items Bought
1	{Laptop, Printer, Tablet, Headset}
2	{Printer, Monitor, Tablet}
3	{Laptop, Printer, Tablet, Headset}
4	{Laptop, Monitor, Tablet, Headset}
5	{Printer, Monitor, Tablet, Headset}
6	{Printer, Tablet, Headset}
7	{Monitor, Tablet}
8	{Laptop, Printer, Monitor}
9	{Laptop, Tablet, Headset}
10	{Printer, Tablet}

a)     

Step 1: Calculating minimum support count :

It’s given that minimum support percentage is 30%, so, the minimum support count is 3

C1

Items	Support Count
{Laptop}	5
{Printer}	7
{Tablet}	9
{Headset}	6
{Monitor}	5

L1

Items	Support Count
{Laptop}	5
{Printer}	7
{Tablet}	9
{Headset}	6
{Monitor}	5

C2

Items	Support Count
{Laptop, Printer}	3
{Laptop, Tablet}	4
{Laptop, Headset}	4
{Laptop, Monitor}	2
{Printer, Tablet}	6
{Printer, Headset}	4
{Printer, Monitor}	3
{Tablet, Headset}	6
{Tablet, Monitor}	4
{Headset, Monitor}	2

L2

Items	Support Count
{Laptop, Printer}	3
{Laptop, Tablet}	4
{Laptop, Headset}	4
{Printer, Tablet}	6
{Printer, Headset}	4
{Printer, Monitor}	3
{Tablet, Headset}	6
{Tablet, Monitor}	4

C3

Items	Support Count
{Laptop, Printer, Tablet}	2
{Laptop, Printer, Headset}	2
{Laptop, Tablet, Headset}	4
{Printer, Monitor, Tablet}	2
{Tablet, Printer, Headset}	4

L3

Items	Support Count
{Laptop, Tablet, Headset}	4
{Tablet, Printer, Headset}	4

C4 will not be generated because one of its subset is not in L3. So, we are stopping here.

Result :

Frequent Item sets

{Laptop}

{Printer}

{Tablet}

{Headset}

{Monitor}

{Laptop, Printer}

{Laptop, Tablet}

{Laptop, Headset}

{Printer, Tablet}

{Printer, Headset}

{Printer, Monitor}

{Tablet, Headset}

{Tablet, Monitor}

{Laptop, Tablet, Headset}

{Tablet, Printer, Headset}

Efficiency of Apriori vs FP Growth Algorithm:

When a database is scanned to check Ck for creating Lk, a large number of transactions will be scanned even they do not contain any k-item set.  For any large application, reading data from disk is really resource consuming one.

For example,  the number of frequent pattern generated for 100 item set is 2^100 -1, which is 1.27×10^30 frequent sub-patterns! It’s a too huge set for any computer to store and compute.

In FP Growth algorithm, it scans database only twice. It’s much more efficient and faster than Apriori algorithm.

1^st time : Scan DB once and find single item frequent pattern

2^nd time : Scan DB again and construct a tree.

Implementation of the above problem using R programming language:

Output: