Recommendation system using unsupervised machine learning algorithm & assoc

International Journal of Engineering Research and Development (IJERD)
ISSN Print: ---------X--------- ISSN Online: ---------X---------
1
Download->http://www.ijerd.net/Journalcureentissue.asp Volume 1, Issue 1
RECOMMENDATION SYSTEM USING UNSUPERVISED MACHINE
LEARNING ALGORITHM & ASSOCIATION RULE MINING
Sunita B Aher
Assistant Professor, Walchand Institute of Technology, Solapur University India
ABSTRACT
Here in this paper we present how the combination of Farthest First clustering &
Apriori association rule algorithm is useful in Course Recommender system which suggest
the best combination of courses based on various student choices collected through Moodle
courses. If we consider only Apriori association rule algorithm to find the result then we need
to prepare the data obtained from Moodle database. But if we consider this combined
approach using clustering & association rule algorithm then there is no need to prepare the
data. So we present in this paper, this new approach & also present the result. To test the
result we have used the open source data mining tool Weka which consists of data mining
task such as classification, clustering & association.
KEY WORDS: Farthest First Clustering Algorithm, Apriori association Rule, Weka, Moodle.
1. INTRODUCTION
The course recommendation system in e-learning is a system that suggests the best
combination of courses in which the students are interested [9].
In this Course Recommendation System, we have considered the 13 course category.
Under each category there will courses. So there are about 82 courses. The framework for
Course Recommender System is given in the figure 1.
Student first logs in the learning management system e.g. Moodle & enrolled for those
courses in which they are interested. This data is stored in the moodle database which we use
to find out the best combination. After collecting the data from student which is stored in
Moodle database, the next stage is to gather & prepare the data. In this step, first we select the
data from database which is relevant. To test the result using Weka i.e. the best combination
courses, first we need to preprocess the data & find out the result. The step-build the model,
INTERNATIONAL JOURNAL OF ENGINEERINGINTERNATIONAL JOURNAL OF ENGINEERINGINTERNATIONAL JOURNAL OF ENGINEERINGINTERNATIONAL JOURNAL OF ENGINEERING
RESEARCH AND DEVELOPMRESEARCH AND DEVELOPMRESEARCH AND DEVELOPMRESEARCH AND DEVELOPMENT (IJERD)ENT (IJERD)ENT (IJERD)ENT (IJERD)
www.ijerd.net
Volume 1, Issue 1 July - August (2013) Pages: 01-11
Research PaperResearch PaperResearch PaperResearch Paper Open AccessOpen AccessOpen AccessOpen Access

2
we directly select the relevant data from Moodle database. After collecting the data from
Moodle database, we clustered the data using clustering algorithm e.g. Simple K-means
algorithm. After clustering data, we use the Apriori Association Rule algorithm to find the
best combination of courses. To find the result we need to preprocess the data from Moodle
database but if we consider the combination of clustering & association rule algorithm then
there is no need to preprocess the data. The preprocessing technique is explained in paper
[10].
Figure 1: Framework for Course Recommendation System using the combination of
Farthest First clustering & Apriori association rule algorithm
2. LITERATURE REVIEW
In paper [2], they explain Aspect oriented programming which offers a unique
module, an aspect, to encapsulate scattered and tangled code, which makes it hopeful to solve
the problem of crosscutting concerns. Identification and encapsulation of crosscutting
concerns is the key problem in the migration from OO system to AO system. A novel aspect
mining method which combines clustering and association rule technology is provided in this
article. Clustering analysis based on the execution traces is provided to find out candidate
aspects; while association rule mining based on the execution traces with ordered call is used
to find out the crosscuts. Both the aspect code (advice body) and the crosscuts (pointcuts) are
gotten after the above two processes, which constitute the aspect mining process.

3
In paper [3], they proposed a method for grouping and summarizing large sets of
association rules according to the items contained in each rule. They use hierarchical
clustering to partition the initial rule set into thematically coherent subsets. This enables the
summarization of the rule set by adequately choosing a representative rule for each subset,
and helps in the interactive exploration of the rule model by the user. Rule clusters can also
be used to infer novel interest measures for the rules.
The main hypothesis discussed in the paper [4] is that Web content analysis can be
used to improve Web usage mining results. They propose a system that integrates Web page
clustering into log file association mining and uses the cluster labels as Web page content
indicators. They experiment with several approaches to content clustering, relying on
keyword and character n-gram based clustering with different distance measures and
parameter settings.
In paper [5] they present a new approach to cluster web images. A tree-distance-based
evaluation measure is used to evaluate the quality of image clustering with respect to
manually generated ground truth. The experiments indicate that combining textual and
content based features results in better clustering as compared to signal only or text-only
approaches.
The main objective of the study [9] was to detect the patterns of use for additional services of
mobile telecommunication through cluster analysis. The authors used empirical data of
mobile telecommunication service collected from a company in Korea. As a result, it was
possible to divide the customers into 8 clusters. After extracting the association rules for the
eight clusters, these are the marketing implications.
The research [11] proposed a bracing approach for increasing web server performance by
analyzing user behavior, in this pre-fetching and prediction is done by pre-processing the user
access log and integrating the three techniques i.e. Clustering, Markov model and association
rules which achieves better web page access prediction accuracy. This work also overcomes
the limitation of path completion i.e. by extracting web site structure paths are completed,
which helps in better prediction, decreasing access time of user and improving web
performance.
The main objective of study [12] was to detect the patterns of use for additional
services of mobile telecommunication through cluster analysis. The authors used empirical
data of mobile telecommunication service collected from a company in Korea. As a result, it
was possible to divide the customers into 8 clusters. After extracting the association rules for
the eight clusters, these are the marketing implications.
3. DATA MINING ALGORITHMS
Here we consider the brief idea about the two data mining algorithm: Clustering
algorithm & Association rule algorithm.
3.1 Clustering Algorithm
Clustering is finding groups of objects such that the objects in one group will be
similar to one another and different from the objects in another group. Clustering can be
considered the most important unsupervised learning technique.

4
3.1.1 Farthest First Clustering algorithm
Farthest first is a variant of K-Means that places each cluster centre in turn at the point
furthermost from the existing cluster centre. This point must lie within the data area. So in
this method, we find the cluster using the farthest first traversal algorithm. This greatly
speeds up the clustering in most of the cases since less reassignment.
3.2. Association Rule Algorithm
Association rules are used to show the relationship between data items. Association
rule generation is usually split up into two separate steps: First, minimum support is applied
to find all frequent itemsets in a database. Second, these frequent itemsets and the minimum
confidence constraint are used to form rules.
3.2.1 Apriori Association Rule
Apriori Association rule is used to mine the frequent patterns in database.
e.g. A person who bought the book Database System Concepts is quite likely also to buy the
book Advanced Database System.
Mining association rules allows finding rules of the form:
If antecedent then (likely) consequent
where antecedent and consequent are itemsets which are sets of one or more items.
Association rule contain the population where population is set of instances.
Support & confidence are two the methods to measures the quality of association rules.
Support is a measure of what fraction of the population satisfies both the antecedent and the
consequent of the rule. Confidence is a measure of how often the consequent is true when the
antecedent is true.
The technique to find the large itemset is:
Pass 1: count support of all sets with just 1 item. Eliminate those items with low support
Pass i: candidates: every set of i items such that all its i-1 item subsets are large
Count support of all candidates.
Stop if there are no candidates [13].
4. RESULT & IMPLEMENTATION
Here we are considering the sample data extracted from Moodle database as shown in
Table 1. Here we consider 45 student & 15 courses. We consider the courses like C-
programming (C), Visual Basic (VB), Active Server Pages (ASP), Computer Network (CN),
Network Engineering (NE), Microprocessor (MP), Computer Organization (CO), Database
Engineering (DBE), Advanced Database System (ADS), Operating System (OS), Distributed
System (DS), Finite State Automata (FSA), Data Structure (DS-I), Software Engineering
(SE), and Software Testing & Quality assurance (STQA). In this table yes represent that the
student is interested in that course & no represent that student do not like that course. If we
apply Apriori association rule algorithm to table 1 then we gets the association rule
containing “no” only. This result is shown in third row of table 6. As we need all the
association rule containing “yes” only for recommending the course to the student, we
prepare the data obtained from Moodle so that we get the correct association rule.
In data preparation stage, we delete those rows & columns from sample table shown in table
1, having very less student count & less course count. After preprocessing of data we got 8

5
courses & 38 rows i.e. 38 students. These 8 courses are C-programming (C), Visual Basic
(VB), Active Server Pages (ASP), Computer Network (CN), Network Engineering (NE),
Operating System (OS), Distributed System (DS), Data Structure (DS-I). The result of data
preparation is shown in table 2. The result of application of Apriori association rule algorithm
on table 2 is shown in fifth row of table 6.
Table 1: Sample data from Moodle database [9]
Courses
Roll_No ↓
C VB ASP CN NE MP CO DBE ADS OS DS FSA DS-I SE STQA
1 yes yes yes yes yes no no no no no no no yes no no
2 no no no no no no no no no no no no no no no
3 yes yes yes yes yes no no no no yes yes yes yes yes yes
4 no no no yes yes no yes no no no no no no no no
5 yes yes yes yes yes no no yes no yes yes no yes no no
6 yes yes yes no no no no no no yes no no yes no no
7 no no no yes yes yes yes no no no no no no yes no
8 no no no no no no no yes yes yes yes no yes no no
9 no no no yes yes yes yes no no no no yes no no no
10 yes no no no no no no no no no no no no no no
11 yes yes yes no no no no no no yes yes no yes no no
12 yes yes yes yes yes no no no no no no no no no no
13 no no no no no no no yes yes yes yes no yes yes yes
14 yes yes yes yes yes no no no no yes yes no no no no
15 yes yes yes no no no no no no no no no yes no no
16 no no no yes yes no no yes yes yes yes no yes no no
17 yes yes yes no no no no no no yes yes no yes yes yes
18 yes yes yes no no no no no no no no no no no no
19 no no no yes yes yes yes yes yes no no no no no no
20 yes no no no no no no no no yes yes no yes yes yes
21 yes no yes no no yes yes no no yes yes yes no no no
23 yes yes yes yes yes yes yes no no yes yes no yes no no
24 yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes
25 no yes yes no no yes yes yes yes yes yes no no no no
28 no no no yes yes no no no no yes yes no yes no no
29 no no no no no yes yes yes yes no no no no no no
30 yes yes yes yes yes no no no no no no no no yes yes
32 yes yes yes no no no no yes yes yes yes no yes no no
35 no no no no no no no no no yes yes no no no no
36 no no no yes yes no no no no no no no yes no no
37 yes yes yes yes yes yes yes yes yes no no no no no no
38 no no no no no no no no no yes yes yes yes yes yes
40 no no no no no no no no no no no no no yes yes
42 no no no yes yes no no no no no no no no no no
43 no no no no no no no no no yes yes no yes no no
44 no no no no no no no no no no no no no no yes

6
Table 2: Sample table 1 after preparation of data
Subject
→
Roll No
of
Student
↓
C VB ASP CN NE OS DS DS-I
1 yes yes yes yes yes no no yes
3 yes yes yes yes yes yes yes yes
4 no no no yes yes no no no
6 yes yes yes no no yes no yes
8 no no no no no yes yes yes
11 yes yes yes no no yes yes yes
12 yes yes yes yes yes no no no
14 yes yes yes yes yes yes yes no
15 yes yes yes no no no no yes
16 no no no yes yes yes yes yes
18 yes yes yes no no no no no
20 yes no no no no yes yes yes
21 yes no yes no no yes yes no
25 no yes yes no no yes yes no
34 yes yes yes no no no no no
35 no no no no no yes yes no
36 no no no yes yes no no yes

7
In the combined approach using Farthest First clustering algorithm & Apriori
association rule algorithm, first we apply Farthest first clustering algorithm on table 1 whose
result is shown in seventh row of table 6. So we clustered the data into three clusters which is
shown in table 3, 4 & 5. Figure 2 shows the visualize cluster assignment window from Weka
tool to cluster the data in three cluster. After saving the file, we will come to know which row
in the belongs to which cluster.
After the clustering the data, we apply the Apriori association rule algorithm on each
cluster & check the result. If the result consist of all association containing “yes” then we
consider the result of that cluster for recommending the course to the student. So out of these
three clusters only cluster 2 gives the correct result shown in table 6.
Figure 2: Visualize cluster assignemnt window in Weka

8
Table 3: Table after application of clustering algorithm- Farthest First algorithm
(Cluster 0)
Courses
Roll_No
|
v
4 no no no yes yes no yes no no no no no no no no
5 yes yes yes yes yes no no yes no yes yes no yes no no
7 no no no yes yes yes yes no no no no no no yes no
9 no no no yes yes yes yes no no no no yes no no no
13 no no no no no no no yes yes yes yes no yes yes yes
16 no no no yes yes no no yes yes yes yes no yes no no
19 no no no yes yes yes yes yes yes no no no no no no
25 no yes yes no no yes yes yes yes yes yes no no no no
29 no no no no no yes yes yes yes no no no no no no
32 yes yes yes no no no no yes yes yes yes no yes no no
35 no no no no no no no no no yes yes no no no no
36 no no no yes yes no no no no no no no yes no no
42 no no no yes yes no no no no no no no no no no
43 no no no no no no no no no yes yes no yes no no
(Cluster 1)
Courses
Roll_No
|
v
1 yes yes yes yes yes no no no no no no no yes no no
6 yes yes yes no no no no no no yes no no yes no no
10 yes no no no no no no no no no no no no no no
14 yes yes yes yes yes no no no no yes yes no no no no
15 yes yes yes no no no no no no no no no yes no no
21 yes no yes no no yes yes no no yes yes yes no no no
30 yes yes yes yes yes no no no no no no no no yes yes
37 yes yes yes yes yes yes yes yes yes no no no no no no
40 no no no no no no no no no no no no no yes yes
44 no no no no no no no no no no no no no no yes

9
(Cluster 2)
Courses
Roll_No
|
v
3 yes yes yes yes yes no no no no yes yes yes yes yes yes
17 yes yes yes no no no no no no yes yes no yes yes yes
20 yes no no no no no no no no yes yes no yes yes yes
23 yes yes yes yes yes yes yes no no yes yes no yes no no
38 no no no no no no no no no yes yes yes yes yes yes
Table 6: Result after application of data mining algorithms
Course
considered
Parameter
Considered Results
Result of Apriori Association Rule before preprocessing & application of combination of Clustering & Association Rule
C, VB,
ASP, CN,
NE, MP,
CO, DBE,
ADS, OS,
DS, FSA,
DS-I, SE,
STQA
Minimum
support: 0.7
Minimum metric
<confidence>:
0.9
Best rules found:
1. Computer_Organization =no → Microprocessor=no
2. Database_Engineering=no → Advanced_Database_System =no
3. Computer_Organization =no Finite_State_Automata =no → Microprocessor =no
4. Microprocessor =no → Computer_Organization =no
5. Software_Testing_&_Quality_Assurance =no → Software_Engineering=no
6. Software_Engineering =no → Software_Testing_&_Quality_Assurance =no
7.Advanced_Database_System=no → Database_Engineering =no
8. Microprocessor =no Finite_State_Automata =no → Computer_Organization =no
9. Finite_State_Automata =no Software_Testing_&_Quality_Assurance =no→
Software_Engineering
=no
10. Finite_State_Automata =no Software_Engineering =no
→Software_Testing_&_Quality_Assurance =no
Result of Apriori Association Rule after preprocessing & before application of combination of Clustering & Association
Rule
C, VB,
ASP, CN,
NE, OS,
DS, DS-I
Minimum
support: 0.5
Minimum metric
<confidence>:
0.9
Best rules found:
1. Distributed_System=yes =yes → Operating_System =yes
2. Visual_Basic =yes → Active_Server_Pages =yes
3. Network_Engineering=yes → Computer_Network=yes
4. Computer_Network =yes → Network_Engineering =yes
5. C_Programming =yes Visual_Basic=yes → Active_Server_Pages =yes
6. Distributed_System=yes =yes Data_Structure-I =yes → Operating_System =yes
7. Operating_System =yes → Distributed_System=yes =yes
8. Active_Server_Pages =yes → C_Programming =yes
9. C_Programming =yes → Active_Server_Pages =yes
10. Active_Server_Pages =yes → Visual_Basic =yes
Minimum
support: 0.6
Minimum metric
<confidence>:
0.9
Best rules found:
1. Distributed_System=yes =yes → Operating_System =yes
2. Operating_System =yes → Distributed_System=yes =yes

10
Result After Application of Clustering algorithm-Simple K-means
C, VB,
ASP, CN,
NE, MP,
CO, DBE,
ADS, OS,
DS, FSA,
DS-I, SE,
STQA
Number of
Cluster:3
Seed:1
Cluster centroids:
Cluster 0
no no no yes yes no no yes yes yes yes no yes no no
Cluster 1
yes yes yes no no no no no no no no no no no no
Cluster 2
yes yes yes yes yes no no no no yes yes yes yes yes yes
=== Model and evaluation on training set ===
Clustered Instances
0 18 ( 40%)
1 20 ( 44%)
2 7 ( 16%)
Result after application of Apriori Association Rule on clustered data
C, VB,
ASP, CN,
NE, MP,
CO, DBE,
ADS, OS,
DS, FSA,
DS-I, SE,
STQA
Minimum
support: 0.85
Minimum metric
<confidence>:
0.9
Number of
cycles
performed: 3
Cluster 0 result(Incorrect result)
Best rules found:
1. Software_Engineering=no 16 → Softare_Tesing_And_Quality_Assurance=no 16 conf:(1)
2. Visual_Basic=no 15 → C_Programming=no 15 conf:(1)
3. Active_Server_Pages=no 15 → C_Programming=no 15 conf:(1)
4. Active_Server_Pages=no 15 → Visual_Basic=no 15 conf:(1)
5. Visual_Basic=no 15 → Active_Server_Pages=no 15 conf:(1)
6. Visual_Basic=no Active_Server_Pages=no 15 → C_Programming=no 15 conf:(1)
7. C_Programming=no Active_Server_Pages=no 15 → Visual_Basic=no 15 conf:(1)
8. C_Programming=no Visual_Basic=no 15 → Active_Server_Pages=no 15 conf:(1)
9. Active_Server_Pages=no 15 → C_Programming=no Visual_Basic=no 15 conf:(1)
10. Visual_Basic=no 15 → C_Programming=no Active_Server_Pages=no 15 conf:(1)
Minimum
support: 0.9
Minimum metric
<confidence>:
0.9
Number of
cycles
performed: 2
Cluster 1 result(Incorrect result)
Best rules found:
1. Advanced_Database_System=no 19 → Database_Engineering=no 19 conf:(1)
2. Database_Engineering=no 19 → Advanced_Database_System=no 19 conf:(1)
3. Computer_Organization=no 18 → Microprocessor=no 18 conf:(1)
4. Microprocessor=no 18 → Computer_Organization=no 18 conf:(1)
5. Microprocessor=no 18 → Database_Engineering=no 18 conf:(1)
6. Microprocessor=no 18 → Advanced_Database_System=no 18 conf:(1)
7. Microprocessor=no 18 → Finite_State_Automata=no 18 conf:(1)
8. Computer_Organization=no 18 → Database_Engineering=no 18 conf:(1)
9. Computer_Organization=no 18 → Advanced_Database_System=no 18 conf:(1)
10. Computer_Organization=no 18 → Finite_State_Automata=no 18 conf:(1)
Minimum
support: 0.95
Minimum metric
<confidence>:
0.9
Number of
cycles
performed: 1
Cluster 2 result (Correct result)
Best rules found:
1. Distributed_System=yes 7 → Operating_System=yes 7 conf:(1)
2. Operating_System=yes 7 → Distributed_System=yes 7 conf:(1)
3. Data_Structure-I=yes 7 → Operating_System=yes 7 conf:(1)
4. Operating_System=yes 7 → Data_Structure-I=yes 7 conf:(1)
5. Data_Structure-I=yes 7 → Distributed_System=yes 7 conf:(1)
6. Distributed_System=yes 7 → Data_Structure-I=yes 7 conf:(1)
7. Distributed_System=yes Data_Structure-I=yes 7 → Operating_System=yes 7 conf:(1)
8. Operating_System=yes Data_Structure-I=yes 7 → Distributed_System=yes 7 conf:(1)
9. Operating_System=yes Distributed_System=yes 7 → Data_Structure-I=yes 7 conf:(1)
10. Data_Structure-I=yes 7 → Operating_System=yes Distributed_System=yes 7 conf:(1)

11
5. CONCLUSION AND FUTURE WORK
Here in this paper we try to present how the combination of data mining technique
such as Farthest First clustering & Apriori association rule algorithm is useful in Course
Recommender system. If we consider only Apriori association rule algorithm & check the
result using weka to find the best combination of courses then we need to prepare the data
obtained from Moodle database. But if we consider this combination of Farthest First
clustering & Apriori association rule algorithm then there is no need to prepare the data. Also
this combination increases the strength of association rule, we obtained. Future works consist
of atomization of this combined approach & checking the result with large amount of data.
6. REFERENCES
[1] Castro, F., Vellido, A., Nebot, A., & Mugica, F. (in press). Applying data mining
techniques to e-learning problems: A survey and state of the art. In L. C. Jain, R. Tedman, &
D. Tedman (Eds.), Evolution of Teaching and learning paradigms in intelligent environment.
Studies in Computational Intelligence (Vol. 62). Springer-Verlag.
[2] Lili He, Hongtao Bai:”Aspect Mining Using Clustering and Association Rule Method”
IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.2A,
February 2006
[3] Al¶³pio Jorge:”Hierarchical Clustering for thematic browsing and summarization of large
sets of Association Rules” Supported by the POSI/SRI/39630/2001/Class Project
[4] Jiayun Guo, Vlado Keˇselj, and Qigang Gao:”Integrating Web Content Clustering into
Web Log Association Rule Mining?” supported by NSERC
[5] Hassan H. Malik, John R. Kender:”Clustering Web Images using Association
Rules,Interestingness Measures,
and Hypergraph Partitions” ICWE’06, July 11–14, 2006, Palo Alto, California, USA. ACM
1-59593-352-2/06/0007.
[7] “Data Mining Introductory and Advanced Topics” by Margaret H. Dunham
[8] Sunita B Aher and Lobo L.M.R.J.. Data Mining in Educational System using WEKA.
IJCA Proceedings on International Conference on Emerging Technology Trends (ICETT)
(3):20-25, 2011. Published by Foundation of Computer Science, New York, USA (ISBN:
978-93-80864-71-13)
[9] Sunita B Aher and Lobo L.M.R.J. Article: A Framework for Recommendation of courses
in E-learning System. International Journal of Computer Applications 35(4):21-28,
December 2011. Published by Foundation of Computer Science, New York, USA ISSN 0975
– 8887
[10] Sunita B Aher and Lobo L.M.R.J.: “Preprocessing Technique for Association Rule
Based Course Recommendation System in E-learning” selected in ICECT-12, proceeding
published by IEEE
[11] Silky Makker, R.K Rathy: “Web Server Performance Optimization using Prediction
Prefetching Engine” International Journal of Computer Applications (0975 – 8887) Volume
23– No.9, June 2011
[12] So Young Sohn, Yoonseong Kim” Searching customer patterns of mobile service using
clustering and quantitative association rule” Published in journal Expert Systems with
Applications: An International Journal, Volume 34 Issue 2, February, 2008
[13] http://www.cse.iitb.ac.in/~sudarsha/db-book/slide-dir/ accessed on 02-03-2012

Recommendation system using unsupervised machine learning algorithm & assoc

Recommended

More Related Content

What's hot (20)

Viewers also liked (9)

Similar to Recommendation system using unsupervised machine learning algorithm & assoc (20)

Recently uploaded (20)

Recommendation system using unsupervised machine learning algorithm & assoc