Download Frequent Pattern Mining with Association Rule Mining

Document related concepts
no text concepts found
Transcript
Presentation
On
An Efficient Approach for Association Rule
Mining by using modified Frequent Pattern
mining Technique
Chintan Bhatt
(Enrollment No. 131150702002)
Guide: Prof. Mehul Patel
Co-Guide: Prof Hitul Patel
Swaminarayan College of
Engineering & Technology
Gujarat Technological
University
OUTLINE
•
•
•
•
•
•
•
•
•
Objective
Introduction
Literature review
Proposed Algorithm
Comparative study
Result and Analysis
Conclusion
Future Scope
References
131150702002
2
OBJECTIVE
Now a days Data Mining is the field in which lots of
researches are carrying on by using a novel
approaches still there is a need to make time and
memory efficient techniques. I am concentrating on
time and memory parameter and also making frequent
pattern mining technique which supports incremental
mining by using modified tree generation method and
compact data structure.
131150702002
3
INTRODUCTION
• Data Mining[10]
– “The nontrivial extraction of implicit, previously unknown,
and potentially useful information”
– Find out useful information from large amount of data.
– Extraction of knowledge from large amount of Data.
• Why Data Mining[10]?
– Too much data and too little information.
– There is a need to extract useful information from the data and
to interpret the data.
131150702002
4
Steps for knowledge discovery in databases
[10]
Database
Data ware house
Data Mining
Patterns
Knowledge
131150702002
5
ARM (Association Rule Mining)[10]






Set of items: I={I1,I2,…,Im}
Transactions: D={t1,t2, …, tn}, tj I
Item set: {Ii1,Ii2, …, Iik}  I
Support of an item set: Percentage of transactions which contain that item set.
Large (Frequent) item set: Item set whose number of occurrences is above a threshold.
Support of AR (s) X  Y: Percentage of transactions that contain X Y
Support = P (X Y)

Confidence of AR (a) X  Y: Ratio of number of transactions that contain X  Y to the
number that contain X
Confidence = P (X Y) / P(X)
Where P is Probability
131150702002
6
Data Mining Techniques [10]
1. Classification
2. Clustering
3. Association Rule Mining
131150702002
7
Literature Survey
131150702002
8
Apriori Algorithm[1][2][3]
• Apriori algorithm is a classical algorithm for
association rule mining.
• It uses Generate and Test approach.
• It is level wise search algorithm where k-itemsets are
used to generate k+1 itemsets.
• It generates candidates itemsets and test whether it is
frequent or not by using min_support and
min_confidence threshold values.
131150702002
9
• Example:TID
List of
Items
D001
C,M,E,A
D002
C,M,E
D003
M,A
D004
C,E
D005
C,E,A
D
MIN_SUPPORT=no. of
occurrences of an item/total no. of transaction.
• First step is to scan database to get the count for each items
call it C1.
• Generate frequent itemsets by using support threshold call it
L1.
• L1 is used to find out L2.
• Do this until Lk.
131150702002
10
TID
List of Items
D001
C,M,E,A
D002
C,M,E
D003
M,A
D004
C,E
D005
C,E,A
Items
C
Support
Count
generate
4
M
3
E
4
A
3
generate
C1
Items
Support
Count
C
4
M
3
E
4
A
3
L1
MIN_SUPPORT=2
• Scan database D again generate C2 by applying L1*L1
relation.
• Generate L2 by using min_support threshold.
131150702002
11
Items Support
Count
C
4
M
3
E
4
A
3
Support
Count
Items
Support
Count
2
(C,M)
2
(C,E)
4
(C,E)
4
(C,A)
2
(C,A)
2
(M,E)
2
(M,E)
2
(M,A)
2
(M,A)
2
(E,A)
2
(E,A)
2
Items
SCAN D
(C,M)
L1*L1
C2
L2
MIN_SUPPORT=2
Items
Items
Support
Count
(C,M,E,A) 1
SCAN D
C4
131150702002
L3*L3
SCAN D
ItemsL2*L2Support
Support
Count
Count
(C,M,E)
2
(C,M,E)
2
(C,E,A)
2
(C,M,A)
1
(C,E,A)
2
(M,E,A)
1
L3
C3
12
• Advantages
– Uses large itemset property
– Easy to implement
• Disadvantages
– Assume transaction database is memory resident.
– Requires many database scans.
– Large no. of candidate generation.
131150702002
13
FP-Growth Algorithm[4][5]
• It overcomes the limitations of Apriori like
huge no. of candidates generation and need to
scan database again and again.
• It uses Divide and conquer approach.
• It compresses the database representing
frequent patterns into a FP-Tree which contain
the itemsets association information.
• Construction of FP-tree uses two pass.
131150702002
14
First Pass
• Take Database D.
• Scan and generate 1-itemsets by sorted frequent items in order
of descending support count.
TID
MIN_SUPPORT=2
List of
Items
D001 C,M,E,A
D002 C,M,E
Items
Find
count
Support
count
Sort
descending
TID
List of
Items
D001 C,E,M,A
D002 C,E,M
C
4
D003 M,A
E
4
D003 M,A
D004 C,E
M
3
D004 C,E
D005 C,E,A
A
3
D005 C,E,A
D
131150702002
L1
15
Second pass
• Construction of FP-tree.
• Create root nod as “null”.
• Use linked list concept and start construction of FP-tree using
support count.
TID
List of
Items
{}
D001 C,E,M,A
D002 C,E,M
D003 M,A
D004 C,E
D005 C,E,A
131150702002
16
Second pass
• Construction of FP-tree.
• Create root nod as “null”.
• Use linked list concept and start construction of FP-tree using
support count.
Items
Sup_
count
C
4
E
4
M
3
A
3
{}
Node
link
C:4
E:4
M:2
M:1
A:1
A:1
A:1
131150702002
17
Mining the FP-tree
• Take initial suffix pattern.
• Construct conditional pattern base.
• "sub-database which contains set of prefix paths in fp-tree
co-occurring with the suffix pattern”
• Perform mining recursively on the tree.
Items
Conditional pattern base
Conditional
FP-tree
Frequent patterns
A
{{C,E,M:1},{C,E:1},{M:1}}
{C:2,E:2}
{C,A:2},{E,A:2},{C,E,A:2}
M
{C,E:2}
{C:2,E:2}
{C,M:2},{E,M:2},{C,E,M:2}
E
{C:4}
{C:4}
{C,E:4}
131150702002
18
• Advantages
– only 2 passes over data-set
– “compresses” data-set
– no candidate generation
– much faster than Apriori
• Disadvantages
– FP-Tree may not fit in memory
– FP-Tree is expensive to build
131150702002
19
ECLAT Algorithm[10]
• It uses TID(transactional id).
• It uses vertical data format(horizontal transactional can be
transformed into the vertical data format).
MIN_SUPPORT=2
TID
List of Items
Itemsets
TID
D001
C,M,E,A
C
D001, D002, D004, D005
D002
C,M,E
D003
M,A
M
D001, D002, D003
D004
C,E
E
D001, D002, D004, D005
D005
C,E,A
A
D001, D003, D005
D
131150702002
VERTICAL DATA FORMAT 1-ITEMSET
20
Items
TID
(C,M)
D001,D002,
(C,E)
D001,D002,D004,
D005
(C,A)
D001,D005
(M,E)
D001,D002
(M,A)
D001,D003
(E,A)
D001,D005
Items
TID
(C,M,E) D001,D002
(C,E,A)
D001,D005
VERTICAL DATA FORMAT 3-ITEMSET
VERTICAL DATA FORMAT 2-ITEMSET
131150702002
21
• Advantage
―Very fast support counting
• Disadvantage
—Intermediate tid-lists may become too large for memory
131150702002
22
CATS Tree algorithm[4][6]
• Compressed and Arranged
Sequences tree algorithm.
• Extension of FP-tree.
• Use only single data scan.
• Contains all elements of FP-tree.
• Supports Interactive mining.
131150702002
Transaction
23
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:1
A:1
C:1
D:1
G:1
I:1
M:1
P:1
131150702002
24
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:1
A:1
A:1
C:1
B:1
D:1
C:1
G:1
F:1
I:1
L:1
M:1
M:1
P:1
131150702002
O:1
25
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:1
A:1
A:1
C:1
B:1
D:1
C:1
G:1
F:1
I:1
L:1
M:1
M:1
P:1
131150702002
Merge common items
O:1
26
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:2
A:2
C:2
B:1
Merge common items
D:1
G:1
I:1
L:1
M:1
M:1
P:1
131150702002
O:1
27
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:2
A:2
C:2
B:1
Swap item
D:1
G:1
I:1
L:1
M:1
M:1
P:1
O:1
131150702002
28
• CATS-tree construction
Root
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
F:2
A:2
C:2
B:1
Add remaining items to
swapped node
M:2
D:1
G:1
L:1
I:1
P:1
131150702002
O:1
29
• CATS-tree construction
TID
Original Transaction
Projected transactions for
FP-tree
1
F,A,C,D,G,I,M,P
F,C,A,M,P
2
A,B,C,F,L,M,O
F,C,A,B,M
3
B,F,H,J,O
F,B
4
B,C,K,S,P
C,B,P
5
A,F,C,E,L,P,M,N
F,C,A,M,P
Root
F:2
A:2
C:2
M:2
B:1
D:1
L:1
G:1
O:1
I:1
P:1
131150702002
30
• Final CATS-tree……
• Frequency of a parent node must be greater than the sum of its children’s
frequencies.
Roo
t
B:1
F:4
B:1
C:1
A:3
H:1
K:1
C:3
J:1
M:3
S:1
P:1
O:1
B:1
P:2
E:1
L:1
D:1
L:1
G:1
O:1
N:1
I:1
131150702002
31
• Sorted frequent item list
C:4 , F:4 , A:3 , B:3 , M:3 , P:3 , L:2 , O:2 ,D:1 , E:1 , G:1 , H:1 , I:1 , J:1 , K:1 ,
N:1 , S:1 Minsup=2
• According to the sorted item list the conditional CATS
tree will be built.
131150702002
32
• Advantages
– Only 1 pass over data-set
– “compresses” data-set
– no candidate generation
• Disadvantages
– Tree construction is expensive to built.
– swapping and/or merging of nodes require extra cost.
– The algorithm needs to traverse both upward and
downward to include frequent items.
131150702002
33
A Time & Memory Efficient Technique for Mining Frequent
Pattern Mining[15]
•
•
•
•
This technique transforms the original data set into a transformed and compacted
data set & then it discovers the frequent patterns from the transformed data set.
Scan the data base (TDB) to find the support count of each single item. Store this
result in a new data structure called Table.
Compare the support of each element of Table to the minimum threshold. If the
support of any element is less then the minimum threshold then that element is
discarded. Now arrange all the elements of Table in the decreasing order of their
support count. Discard all the infrequent item found in step2 are discarded from the
original TDB. In this way, we will get a new NTDB, whose transaction will contain
elements with support count greater than the threshold. Now rearrange all the
transactions of NTDB in the decreasing order of their item count.
Store all the transactions and their count in a multidimensional table (MTable).
Then select transaction of highest size whose count is greater than the minimum
threshold. If no such transaction found then select highest sized and second highest
sized transaction to generate the second highest sized item set. Continue this
process until frequent item sets with greater support count are found.
131150702002
34
Proposed Algorithm
which supports….Incremental Data Mining and also…..
•
Frequent pattern mining without generation of candidate
item sets.
•
Enable frequent pattern mining with different support
without rebuilding the tree structure.
•
Allow mining with a single pass over the database as well as
insertion and deletion of transactions at any time.
131150702002
35
INPUT
DB (Database)
Min_support
Step 1:
Scan DB and count number of each item set and discard item which does not
support minimum threshold value call it NDB.
Min_support=3
TID
List of Items
D001
C,M,E,A,P
D002
C,M,E
C
4
D003
M,A,F
E
4
D004
C,E,P
M
3
D005
C,E,A
A
3
P
2
F
1
DB
131150702002
Items
Support
count
TID
List of Items
D001
C,M,E,A
D002
C,M,E
D003
M,A
D004
C,E
D005
C,E,A
NDB
36
Step 2:
• Store discarded items in Temp Array.
Items
Support
count
P
2
F
1
Temp Array
131150702002
37
Step 3:
• Arrange items in alphabetical order in NDB.
TID
List of Items
TID
List of Items
D001
C,M,E,A
D001 A,C,E,M
D002
C,M,E
D002 C,E,M
D003
M,A
D003 A,M
D004
C,E
D004 C,E
D005
C,E,A
D005 A,C,E
NDB
131150702002
38
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
List of
Items
{}
D001 A,C,E,M
D002 C,E,M
D003 A,M
D004 C,E
D005 A,C,E
131150702002
39
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
D002 C,E,M
D003 A,M
D004 C,E
D005 A,C,E
131150702002
40
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
D002 C,E,M
D003 A,M
C:1
D004 C,E
D005 A,C,E
131150702002
41
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
D005 A,C,E
131150702002
E:1
42
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
131150702002
43
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
C:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
131150702002
44
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
C:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
E:1
D005 A,C,E
M:1
131150702002
45
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:1
C:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
E:1
M:1
M:1
131150702002
46
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:2
C:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
E:1
M:1
M:1
131150702002
47
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:2
C:1
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
E:1
M:1
M:1
131150702002
48
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:2
C:2
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
E:1
M:1
M:1
131150702002
49
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:2
C:2
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
E:2
M:1
M:1
131150702002
50
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:3
C:2
D002 C,E,M
C:1
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
E:2
M:1
M:1
131150702002
51
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:3
C:2
D002 C,E,M
C:2
D003 A,M
D004 C,E
E:1
D005 A,C,E
M:1
E:2
M:1
M:1
131150702002
52
Step 4: Construction of
tree.
Step 5. Create root nod as “null”.
Step 6. Use linked list concept and start construction of tree using
support count.
TID
{}
List of
Items
D001 A,C,E,M
A:3
C:2
D002 C,E,M
C:2
D003 A,M
D004 C,E
E:2
D005 A,C,E
M:1
E:2
M:1
M:1
131150702002
53
Step 7:
Mine tree by using FP-growth mining technique.
•
Step 1: The Conditional pattern base will be formed according to the ascending
order of the items.
•
Step 2: Conditional FP-tree will be generated according to the same order as in step
1 by removing the items with the frequency less than the min_sup from the
conditional pattern base.
•
Step 3: Finally frequent patterns will be generated from the conditional FP-tree.
131150702002
54
STEP 8:
Update database by inserting new transaction. Count all items if any item available
in temp array increment its count and then check all items count with Min_support
value.
Items
Support
count
D006 A,P,F
C
1
D007 C,P
A
1
P
2
F
1
TID
List of
Items
P satisfy Min_support value 3 so add new
transaction in NDB then follow all steps.
131150702002
Increment same
item in Temp Array
Items
Support
count
P
4
F
2
Temp Array
55
COMPARATIVE STUDY
Properties
Apriori
FP-Growth
ECLAT
CATS
Extension of FPgrowth
Number of scans required in
the best case
2
2
2
1
1
Number of scans required in
the worst case
K+1
2
K+1
1
1
Candidate generation
Yes
(is the
No
Yes
No
No
bottleneck)
Interactive Mining
No
No
No
Yes
Yes
Incremental mining
No
No
No
Partially but
Unclear
Yes
Execution time
slow
Fast over
Apriori
Fast
Fast
Will test after
implementation
Memory
Large
Large than
Apriori
Large than
Apriori & FPGrowth
Large
Will test after
implementation
131150702002
56
Result and Analysis
131150702002
57
•
•
The goal of experiment is to find out the performance of proposed algorithm over
existing algorithms.
First experiment shows comparison between Apriori, FP-growth and Proposed
Algorithm by requiring time for different min_sup value. Results show that
Proposed Algorithm requires minimum time as compared to Apriori and FP-growth.
The Apriori algorithm works on the principle of candidate generate and test, so it
requires the maximum execution time.
First Experimental result
T
I
m
e
(
s
e
o
n
d
)
2.6
2.4
2.2
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Apriori
FP-growth
Proposed Algorithm
20%
30%
40%
50%
Min_sup
131150702002
58
•
Second experiment shows that whenever there is an update in database FP-growth
algorithm requires more memory because there is a need to build tree from the start
but in Proposed Algorithm it requires less memory than FP-growth because it
allows incremental mining.
Second Experimental result
6
5.5
5
4.5
M
4
e
m 3.5
o
3
r
y 2.5
FP-growth
Proposed Algorithm
(
M
B
2
)
1.5
1
0.5
0
5k
10k
15k
20k
Dataset
131150702002
59
•
Third experiment shows that mining for different min_sup FP-growth algorithm
requires more time because there is a need to build tree from the start but in
Proposed Algorithm it requires less time than FP-growth because it allows iterative
mining.
Third Experiment result with different min_sup
0.4
T
i 0.3
m
e
(
s
0.2
e
c
o
n 0.1
d
FP-growth
Proposed Algorithm
)
0
5%
10%
15%
20%
25%
30%
35%
40%
min_sup
131150702002
60
•
Finally, all experiments show that if any modifications are proposed, as per the
algorithm of FP-Growth, the tree generation procedure has to be started from the
scratch.
•
In the Proposed Algorithm, if any transaction is going to be added, inserted or
deleted there is a provision to make changes directly in the existing tree because it
uses alphabetical order. So for incremental size of the database, Proposed
Algorithm is better than any of the existing algorithms.
131150702002
61
Min_sup
Apriori
FP-growth
Proposed Algorithm
Time in second
20%
30%
2.3
1.6
0.2
0.17
0.21
0.15
40%
50%
1.23
1.11
0.25
0.2
0.15
0.19
First Experiment Table
Dataset
FP-growth
Proposed Algorithm
Memory in MB
131150702002
5k
4.71
5.59
10k
1.17
0.65
15k
3.71
2.8
20k
1.9
Second Experiment Table
0.72
62
Min_sup
FP-Growth
Proposed Algorithm
Time in second
5%
0.281
0.312
10%
0.156
0.265
15%
0.125
0.125
20%
0.125
0.109
25%
0.219
0.156
30%
0.14
0.109
35%
0.109
0.109
40%
0.125
0.093
Third experiment result Table
131150702002
63
Conclusion
• Modification done in FP-growth algorithm captures transactions of
database and arranges nodes according to alphabetical order that is
unaffected by changes in item frequency it also compact the
database by using temp array.
• By exploiting its nice properties, proposed algorithm can be easily
maintained when there is an update in database transactions.
• Proposed algorithm does not require merging and/or splitting of tree
nodes.
• It avoids the rescan of the entire updated database or the
construction of a tree from the scratch for incremental updating.
131150702002
64
Future Scope
• Still there is a need to reduce execution time and required memory for
frequent pattern mining technique.
• Incremental mining can be enabled without using alphabetical order by
changing in data structure.
• Several graph based techniques, finite automata (stack manipulation) and
ant colony (ACO) optimization can be used for frequent pattern mining.
• Frequent pattern mining can be done by using Neural Network Technique.
131150702002
65
References
[1] Agrawal R, Imielinski T, Swami AN. "Mining Association Rules between Sets of Items
in Large Databases." SIGMOD. June 1993.
[2] R. Agrawal and R. Srikant, “ Fast algorithms for mining association rules”, Proceeding
of the 20th VLDB Conference Santiago, Chile 1994.
[3] R Agrawal, Mannila H, Toivonen H, Verkamo AI.“Fast Discovery of Association Rules." at
Quest Project at IBM Almaden Research Centre and research at the university of
Helsinki 1994.
[4] Cheung W., ”Frequent Pattern mining without candidate generation or support
constraint.” Master’s thesis, University of Alberta, 2002.
[5] Jiawei Han, Jian Pei, and Yiwen Yin,” Mining Frequent Patterns without Candidate
Generation “, Simon Fraser University, 2002.
[6] William Cheung and Osmar R. Zaiane, “Incremental Mining of Frequent Patterns
without candidate Generation or Support Constraint”, IDEAS’03.
[7] Christian Borgelt, “ An Implementation of the FP-growth Algorithm” OSDM’05.
[8] Q. I. Khan, T. Hoque and C. K. Leung, “ CANTree: A Tree structure for Efficient
Incremental mining of frequent patterns”, ICDM ’05.
[9] Sanjay Patel and Dr. Ketan Kotecha, “Incremental Frequent Pattern Mining using Graph based
approach”, International Journal of Computers & Technology, March-April 2013.
[10] Jiawei Han and Micheline Kamber, Book.”Data Mining, Concept and Techniques”.
131150702002
66
[11] “An Automata Approach to Pattern Collections”, Taneli Mielik¨ainen HIIT Basic
Research
Unit
Department
of
Computer
Science
University
of
Helsinki,
Finland
[email protected]
[12] Jayanta Kumar Basu,Debnath Bhattacharyya, Tai-hoon Kim, “Use of Artificial Neural
Network in Pattern Recognition “, International Journal of Software Engineering and Its
Applications Vol. 4 No. 2 April 2010.
[13] Ketki Muzumdar, Ravi Mante, Prashant Chatur,” Neural Network Approach for Web Usage
Mining”, International Journal of Recent Technology and Engineering (IJRTE) ISSN: 22773878, Volume-2, Issue-2, May 2013
[14] Mark W Craven, Jude W Shavlik,” Using Neural Networks for Data Mining”, Submitted to
the Future Generation Computer Systems special issue on Data Mining.
[15] Pradeep Rupayla, Kamlesh Patidar “A Time & Memory Efficient Technique for Mining
Frequent Pattern Mining ” International Journal of Innovative Research in Computer and
Communication Engineering ISSN(Online): 2320-9801 ISSN (Print): 2320-9798, Vol. 3, Issue
2, February 2015
[16] A.Meenakshi “SURVEY OF FREQUENT PATTERN MINING ALGORITHMS IN
HORIZONTAL AND VERTICAL DATA LAYOUTS” International Journal of Advances in
Computer Science and Technology, ISSN 2320 – 2602 Volume 4 No.4, April 2015
WEBSITE
[17] http://fimi.cs.helsinki.fi/data
[18] http://www.almaden.ibm.com/cs/quest//syndata.html#assocSynData
131150702002
67
Related documents