Associative Rule Mining Video Lecture | What is Data Analytics? - Data & Analytics

hello and welcome to this module on
Association rule mining or frequent
pattern mining which is essentially the
basic problem underlying Association
rule mining so we had a very brief look
at Association rule mining it's very
beginning of the machine learning
section right so the idea behind
Association rule mining is to first mind
frequent patterns that occur in the data
and based on the frequent patterns that
we have mined derive Association rules
which is the form that if a happens then
B is likely to happen right so it
basically is like a conditional
dependence relation that you are mining
if a happens that makes me more likely
it happened you could find such patterns
in sequences looking at time series data
like financial data or looking at fault
analysis where one thing causes a fault
occur or you can look at in the
transactional data column in a context
which is where workers originally
proposed and that is what we will look
at in more detail in the rest of the
module and more interestingly you could
also look at mining frequent patterns
and associations and graphs which is
probably used in social network analysis
so let us look at the mining
transactions for the rest of the module
so the transaction is a collection of
items that were bought together right
that's a the simple definition that we
will use for the purposes of Association
rule mining and please note that a set
or a subset of items is usually denoted
an item set in the Association rule
mining community so the goal here is to
find first to find frequent itemsets
then you would say that an item set a
implies item set B for example if you
could say that if somebody buys milk
then they are likely to buy bread if
both a and the event a union B
or frequent itemsets that would mean
that both a honey which is milk in this
case and a union B which is bread and
milk both should be frequent in which
case I can say that if you buy milk then
you buy bread as well let's take a look
at it's a simple example here so here is
a set of transactions so I have five
transactions and each color here denotes
a different kind of fighter alright so
the first three transactions are four
items it's the fourth one is a three
item set and the fifth one is a two item
set and let's assume that you have a
frequency threshold of three right so we
essentially have the following as
frequent one item sets so you have blue
which occurs in all the five
transactions and then purple which
occurs in four transactions and pink
which occurs the pink item checkers in
three transactions so none of the other
items occur in three or more
transactions so the frequent one item
sets are just these so the next thing
you have to look at is the frequent two
item sets the frequent two item sets you
can see or essentially powerful and blue
which occur in four transactions and
pink and blue which occur in three
transactions right so none of the other
combinations are frequent so the only
other thing which we really have to look
at this purple and pink and purple and
pink occurred only in two of the
transactions together therefore they are
not frequent so the goal here is to
first find such frequent itemsets and
from these frequent itemsets how do we
determine which are interesting
association rules so the two measures of
interestingness for association rules
are essentially support so the support
of the rule is the percentage of item
sets that contain a union B right so in
this case and then the confidence of a
rule is the other measure that we are
interested she'll go back and look at
the data set once we have understood
what support and confidences so the
confidence of a rule is the percentage
of item sets containing a that also
contained a union B right so essentially
that tells you how confident you are in
making this
right so typically we look for roles
with both high support and confidence
right if you think about it in both in
the case of support and confidence we
really need to find thee the frequency
of the item sets right so once we
determine what are the frequent itemsets
and what their frequencies are then we
can easily determine what are the
relevant Association rules so more
effort needs to be focused on counting
more rather than the Association rule
itself so that is why I said the
frequent pattern binding part of this
more interesting than the Association
rule mining part right let's go back and
look at the patterns that we have mined
so far so if you remember the one item
sets aren't really interesting except to
establish the frequency part of it so
let's look at a rule which says that
purple implies blue that if you have
bought purple then you are likely to buy
do right so our purple occurs in your
transactions and blue is likely to occur
so if you think about it this is a valid
rule to have because both purple as well
as purple and blue where frequent right
in in the earlier slide we saw that both
purple and purple into a frequency so it
satisfies the a a union B rule and what
about the support of this rule the
support is essentially all the
transactions the number of transactions
in which both a and B occur so that we
are very a union B occurs divided by the
total number of transactions so a union
B occurs in four-fifths of this data set
and therefore the support of this rule
is four by five what about the
confidence of the rule the Conference of
the rule is whenever a occur right how
many times a union B is occurred and in
this case it turns out that whenever a
across a union B occurs their fourth
conference is one right so likewise for
the pink implies blue rule you can see
that this support is 3 by 5 because that
is how many times pink Union blue occurs
and the conference is one because
whenever pink occurs think Union blue
also occurs right so what about this
rule I mean is this a valid rule to look
all right both blue is frequent and drew
in purple is also frequent so that is a
valid room to look at but is that a
better rule than the one that we saw
earlier that's not necessarily the case
because the support here is four by five
and the confidence is four by five as
well so the earlier rule which is purple
implies blue had a higher confidence
than this rule but both of these are
possible rules that you could consider
because both the rules have a high
support and a high confidence so this is
essentially the idea behind Association
rule mining right so it can be applied
to a variety of applications market
basket analysis is one so as I had
mentioned earlier so market basket
refers to the fact that when you go to a
supermarket you're going to buy a set of
items together and put them together in
your basket so essentially you're
looking at what goes into the basket at
the market right so this these baskets
are considered as transactions and you
look at frequent pattern mining in these
transactions could look at co-occurrence
of words right and that can be used to
derive certain kinds of topic
relationships
all right people have looked at
plagiarism detection in turn in terms of
frequent pattern mining now people have
applied this to biological data looking
at biomarkers genes or proteins versus
diseases right so trying to find out
associations between co-occurrence of
certain protein abnormalities and
diseases also looked at within the
context of time series where
co-occurrence of events can be used to
model trigger events English scientific
trade agreements okay so that brings us
to the end of the first module on
frequent pattern mining and the
subsequent module will look at
techniques for efficiently mining
frequent patterns