See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/293464737

main steps for doing data mining project using weka

Technical Report · February 2016

CITATION READS
1 13,792

1 author:

Dalia Sami Jasim

Universiti Kebangsaan Malaysia
9 PUBLICATIONS   36 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Timetable Scheduling For Secondary Schools in Iraq View project

All content following this page was uploaded by Dalia Sami Jasim on 08 February 2016.

The user has requested enhancement of the downloaded file.

 Data Mining Approach and Its Application
to Dresses Sales Recommendation

By

Dalia Sami Jasim

Abstract:

The fundament of data mining (DM) is to analyses data from various points of view. Classify the
data and summarize it, DM has begun to be widespread in every and each application. Although
we have huge magnitude of data, but we do not have helpful information in each field, there are a
lot of DM software and tools to aid us the advantageous information. In this work we give the
essentials of DM steps such as preprocessing data (remove outlier, replace missing values etc.),
attribute selection, aim to choose just relevant attribute and removing the irrelevant attribute and
redundant attribute, classification and assessment of varied classifier models using WEKA tool.
The WEKA software is helpful for a lot of application’s type, and it can be used in different
applications. This tool is consisting of a lot of algorithms for attribute selection, classification
regression and clustering.

Keywords
Data mining, Weka, preprocessing, classification.
1. Introduction :

Data mining (DM) or knowledge discovery is the procedure of using statistical techniques and
knowledge-based methods to analyze the data to mine patterns having meaning from vast data
sets and change these into helpful information. Through DM process, diverse techniques are
utilized to detection relationships, patterns or associations among the dataset features, that can be
transform into knowledge around past patterns and trends of future. In general whatever four
types of relationships are discussing: classes, clusters, sequential and association’s patterns.
Classification category deals with DM techniques that looking for cluster and class
relationships [1].
Classification point out the DM side that making an attempt to predict to which class every
observation of the dataset must be placed by constructing a model based on some predictor
attributes. Classification methods divided into: unsupervised or supervised. In supervised
classification, one attribute of the dataset include predetermined values that represent a collection
of the data. These collections are called classes. For unsupervised classification the objective is
to partition into groups or clusters the observations of the dataset based on some logical
relationship that exists among the values of the attributes but that must yet be discovered [2].
And to achieve classification we firstly need to preprocess the data that we already takes from
UCI web page and to do that we will use Weka to achieve all data mining process. Weka tool is
software for data mining existing below the General public license (GNU). The Weka system is
developed at Waikato University in New Zealand. Weka is available for free at
http://www.cs.waikato.ac.nz/ml/weka. The Developers used Java to write this system. Weka
provides implementation of data mining and machine learning algorithms. User can achieve
classification, association, clustering, filtering, regression, visualization etc. by using Weka tool.
The focus of this report is to use an existing dataset (Dresses_Attribute_Sales Data Set for year
2014) from UCI Machine Learning Repository to preprocessing data for data mining process.
This dataset contain Attributes of dresses and their recommendations according to their sales.
Sales are monitor on the basis of alternate days. There are many preprocessing data techniques
like (clean the data, reduce the size of data and transform data into appropriate type).then we can
use this data set for a Recommender System. The objective of a Recommender System is to build
up meaningful recommendations for users about items or products that might interest them.

2. Related literature on using the dataset:

Until now no one use dresses sales dataset from UCI, but there are several approaches have been
proposed in the field of recommendation system using data mining approach such as [3] that
proposed medical advices recommendation system based on a hybrid method using varied
classifications and unified Collaborative Filtering. Multiple classifications based on decision tree
algorithms are applied to build an accurate predictive model that predicts the disease risk
diagnosis for the monitored cases. [4] Designed a novel book recommendation system. Readers
will be redirected to the recommendation pages when they can’t find the required book through
the library bibliographic retrieval system. The recommendation pages contain all the essential
and expanding book information for readers to refer to. Readers can recommend a book on these
pages, and the recommendation data analyzed by the recommendation system to make scientific
purchasing decision, the author proposed two formulas to compute the book value and copy
number respectively based on the recommendation data. In same trend [5] Presented a
recommendation technique based on opinion mining to propose top ranked books on different
discipline of the computer science. Based on the need of the customers and the reviews collected
from them, they have categorized features for the books and analyze the features on the basis of
several characteristics that they had categorized and reviews of the users. Assigned Weights to
categorized features according to their importance and usage, and accordingly the ranks are
given. Finally, top ten ranked books are listed.
[6] Proposed a movie recommendation system that has the ability to recommend movies to a new
user as well as the others. It mines movie databases to collect all the important information, such
as, popularity and attractiveness, required for recommendation. It generates movie swarms not
only convenient for movie producer to plan a new movie but also useful for movie
recommendation. [7] Introduced a different approach to recommender system which learn rules
for user preferences using classification based on Decision Lists. they had followed two Decision
List based classification algorithms like Repeated Incremental Pruning to Produce Error
Reduction and Predictive Rule Mining, for learning rules for users past behaviors. The authors
also list out their proposed recommendation algorithm and discuss the advantages as well as
disadvantages of our approach to recommender system with the traditional approaches. They had
validated their recommender system with the movie lens data set that contains hundred thousand
movie ratings from different users, which is the bench mark dataset for recommender system
testing.

2.1 PRELIMINARY ON DATA DESCRIPTION

2.2 DATA DESCRIPTION
The dataset represents dresses sales and the recommendations according to this sales. It is
include 13 features and class representing Price, Size, Season, NeckLine , waiseline, Material,
FabricType, Rating, Decoration, SleeveLength ,Style, Pattern, Type, Recommendation .
2.3 Source of Data: This dataset produced by Muhammad Usman & Adeel Ahmed, and
Students at Air University.
2.4 Description on Domain problems: The domain problem we can use this data set with it is
recommendation system that try to recommend products to customers depend on ratings or the
behavior of past customer.
2.5 List of attributes with their types:

ID Attribute Type
1- Dress_ID numeric
2- Style categorical
3- Price categorical
4- Rating numeric
5- Size categorical
6- Season categorical
7- NeckLine categorical
8- SleeveLength categorical
9- waiseline categorical
10- Material categorical
11- FabricType categorical
12- Decoration categorical
Figure 1.Attribute visualization
13- PatternType categorical
14- Recommendat numeric
ion
Tabel1. List of attributes

Figure 2. DataSet description

3. Preliminary examination of data:

3.1 Are there any “obvious” patterns in the data which might be helpful?
No, I can’t see any obvious patterns in the dataset.

3.2 Are there any experts that understand the data well and with whom you can talk?

No, I don’t know any experts that know the data and I can speak with him.
3.3 Are values missing for some attributes?
Yes, there are a missing values in attributes (price , season, neckline , waiseline , material
, FabricType ,decoration , PatternType) .

Figure 3. Show Missing values in gray box

3.4 Examination of dataset using statistical methods:

When we load the dataset in Weka, in the right side we can see information about the
selected attribute, like its value and how many times an instance in the dataset has a
particular value, mean and Standard Deviation matures for numeric values.And the class
attribute is numeric so to achieve classification requirements we must change the class
from numeric to nominal using the path:
Choosefiltersunsupervised attribute  NumericToNominal
Also with this dataset we have ID attribute ,so we have to delete this attribute because it
is full discrete and not give any benefit to classification .

4. Material and Methods: Describe the preprocessing technique used to each attribute in the
dataset:
1.2

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25
-0.2

Figure 4 sample of raw data

 4.1 Data cleaning: my dataset (Dresses Sales) need to clean from these problems:
 Incomplete: lacking attribute values, missing values within 4 records, so we can delete
these records.
 inconsistence : containing discrepancies in codes or names, in this dataset we have
duplicate attribute values :

Figure 5. example inconsistence name of attribute

So we can handle this problem using notepad++ by follow the path Search  Replace

Figure 6. Replace window in notepad++

After click on Replace all the result is:

Now we still have same problem with Autumn and it is duplication value Automn, Spring and
it is duplication value spring, and Winter with its duplication value winter, so we can handle
this problem with same procedure as above, and the result by checking with Weka is :

Figure 7. Attribute Season before remove duplicate names Figure 8. Attribute Season after remove duplicate names
 We have same problem with attributes, NeckLine(Sweetheart- sweetheart), Style(Sexy- sexy),
price(Low-low,High-high),size(small,s,S),sleevelength(sleeveless-sleevless-sleeevless
sleeveless , cap_sleeves - capsleeves , halesleeve - half ), Meterial (chiffon - chiffonfabric) ,
so we can solve this problem with same procedure.

 noisy: containing errors or outliers : to check if we have any outliers in our dataset
(Dresses Sales)we use the filter InterquartileRange , this filter adds new attributes that
point out whether values of instances can be considered outliers or extreme values.
Choose filters  unsupervised attribute InterquartileRange then click apply and
the result is:

Figure 9. the Outliers after implement IQR filter on the dataset

As we show in figure above the filter add new attributes (outlier and extreme values) so referring
to this result we have 120 outlier in our dataset, and about extreme values the result is zero so we
don’t have any extreme values.Now to remove outlier values we can use the path :

Choose filters  unsupervised instanceRemoveWithValues set the properties of

attributIndes to 14(the index of outlier attribute) and nominalIndices to last(to remove just the
last values that have the value yes) and after applying the result is:
 Figure 10. Removing the Outliers 1

4.2 Data reduction: get a new data representation that is considerably smaller in magnitude but
so far produces the same (or roughly same) analytical results. Data reduction strategies:

 Numerosity reduction : we can be applied Numerosity reduction technique to decrease the

data volume by choosing alternative, smaller forms of data representation

Here we change the attribute values from nominal to numeric, as example:

Attribute season: Summer, Autumn, Spring, Winter  Change to: 1,2,3,4 And so on for all
nominal values.

Some values take a large scale as example attribute material from 0-23. So to reduce this scale
we use the filter discretizes using:
Style 126 NeckLine 168 SleeveLength 115 Material 2410 FabricType 2210
Decoration 2412 PatternType 147
Choose Filters  unsupervisedattributediscretize 

Figure 11. Example of attributes before discretization Figure 12. Example of attributes after discretization

And so on for all high scale attribute.

 Dimensionality reduction: Dimensionality reduction reduces the dataset volume by removing
irrelevant attributes.

 Here we use the filter AttributeSelection:

ChoosefilterssupervizedattributeAttributeSelection And the result is:

Figure 13. Attributes after applying attribute selection

5 Classification: Classification is a machine learning and Data mining technique applied to

predict membership for a particular group of data instances. It’s the problem of finding the model
for class attribute as a function of the values of other attributes and predicting an accurate class
assignment for test dataset. And the classification algorithms we will use in this work are:

5.1 Decision tree:

Decision tree can be used as a model for sequential decision problems under uncertainty. A
decision tree describes graphically the decisions to be made, the events that may occur, and the
outcomes associated with combinations of decisions and events. Probabilities are assigned to the
events, and values are determined for each outcome. A major goal of the analysis is to determine
the best decisions. decision tree advantages are (it is easy to fit, easy to employ, and easy to
interpret as a fixed sequence of simple tests, It is non-linear, so It is work much better than linear
models for highly non-linear functions).and disadvantage of decision tree is it can be extremely
sensitive to small confusion in the data: a small change can result in a Significantly different tree.
It can easily overfit. This can be eliminating by validation methods and pruning, but this is a grey
area. It can have problems out-of-sample prediction. [8] introduced an extensions to the Decision
tree construction which enhance the efficiency and utility of the method for a movie
recommendation service, and implemented decision tree learning algorithm on the Netflix
training data (100Mratings), and then evaluated its performance on the Netflix test data (2.8M
ratings). [9] Described a case study of the exploitation of Decision Trees for creating an
industrial recommender system. The aim of this system is to recommend items of a fashion retail
store chain in Spain, producing leaflets for loyal customers announcing new products that they
are likely to want to purchase.
[3] Proposed medical advices recommendation system based on a hybrid method using multiple
classifications and unified Collaborative Filtering. Multiple classifications based on decision tree
algorithms are applied to build an accurate predictive model that predicts the disease risk
diagnosis for the monitored cases.
[10] Investigated the assumption that decision trees perform better than twenty other
classification algorithms in classifying binary data by comparing the decision trees with a fuzzy
set-based classifier and the naive Bayes on real and artificial datasets. And in the field of job
recommender systems[11] presented recommend jobs to the candidates according to their
matching profiles using decision tree where In these each internal node represents a condition,
every edge coming out represents the choice and every leaf node represents the classification or
the decision.

5.2 Naïve Bayes

The Naive Bayesian (NB) classifier is based on Bayes’ theorem with independence assumptions
between predictors. A Naive Bayesian model is easy to build, with no complicated iterative
parameter estimation which makes it particularly useful for very large datasets. Despite its
simplicity, the Naive Bayesian classifier often does surprisingly well and is widely used because
it often outperforms more sophisticated classification methods. An advantage of NB classifier is
that it only needs a small size of training data to assessment the parameters (means and variances
of the variables) necessary for classification. Because independent variables are supposed, only
the variances of the variables for each class need to be determined and not the entire covariance
matrix. The disadvantage is the conditional independence of class assumption by the Naïve
Bayes is not always true, thus leading to low accuracy in some cases. [12] Presented a unique
switching hybrid recommendation approach by combining a Naive Bayes classification approach
with the collaborative filtering. Experimental results on two different datasets, show that the
proposed algorithm is scalable and provided better performance in terms of accuracy and
coverage–than other algorithms while at the same time eliminates some recorded problems with
the recommender systems, in the same area [13] proposed a Naïve Bayes Classifier technique
that uses the user rating for the item and item information to construct a model with time
awareness. Naïve Bayes classifier was used to construct a user model by combining user-item
rating, item feature's information and time information, which improve recommendation
accuracy and can provide an item to user preference at that time which can satisfy users more.
[14]presented content based methods for recommendation systems which are mainly based on
naïve Bayesian machine learning algorithm, presented a study of techniques which suggest naïve
Bayesian algorithm for similarity in recommendation systems.[15] proposes a book
recommendation technique based on opinion mining and Naïve Bayes classifier to recommend
top ranking books to buyers . the author extracted, summarize and categorize all the customer
reviews of a book, also considered the important factor like price of the book while
recommendation and presented a novel tabular efficient method for recommending books to the
buyer, especially when the buyer is coming first time to the website. [16] Presented a Naive
Bayes Classifier Weighing Technique that applied to use with Singular Value Decomposition
Technique for solving sparsity problem. The comparison Mean Absolute Error between Hybrid
Naive Bayes Classifier Weighing and Singular Value Decomposition Technique and Pure
Singular Value Decomposition Technique that found Hybrid Naive Bayes Classifier Weighing
and Singular Value Decomposition yield lower Mean Absolute Error than Singular Value
Decomposition.
5.3 Artificial Neural Network
Artificial Neural Network (ANN) get inspired from the biological nervous systems way, like the
brain, process information. The main ingredient of this algorithm is the unfamiliar structure of
the information processing system. It is consists of a large number of strongly interconnected
processing elements (neurons) working in harmony to solve particular problems. ANNs, similar
to human, learn by example. An ANN is configured for a specific application, such as pattern
recognition or data classification, through a learning process. Learning in biological systems
involves adjustments to the synaptic connections that exist between the neurons. This is true of
ANNs as well.
Some advantages of neural networks:
1. More like a real nervous system.
2. Parallel organization permits solutions to problems where multiple constraints must be
satisfied simultaneously.
3. Graceful degradation.
4. Rules are implicit rather than explicit.
A disadvantage:
1. Difficulty with infinite recursion and structured representations.
Many researchers using ANN in their works like [17] that proposed a neural network hybrid
recommender system have the ability to supply customers, associated with XML-based personal
agents within a multi-agent system called MARF, with suggestions about flights purchases.
MARF agents continuously monitor customers' interests and preferences in their commercial
Web activities, by constructing and automatically maintaining their profiles. In order to highlight
the benefits provided by the proposed flight recommender. In [18] a combination of content-
based, model and memory-based collaborative filtering techniques is used in order to remove the
drawbacks of Content based recommender systems, and to present predicted ratings more
accurately. The training of the data is done using feedforward backpropagation neural network
and the system performance is analyzed under various circumstances like number of users, their
ratings and system model. [19] Studied Artificial Neural Network and optimized it based on
classification system. The performance of the ANN based on classification system was evaluated
by varying number of generations and computing accuracy at different factors. Three standard
data had been used to computer the accuracy. [20] Presented road detector that extracts curb and
navigable surface information from a multilayer laser sensor data. The road data was trained with
an artificial neural network and classified into eight road geometries: straight road, right turn, left
turn, right side road, left side road, T intersection, Y intersection and crossroad. The main
advantage of this method is its robustness to light variations for detecting distinct roads even in
the presence of noisy data thanks to the ANN. [21] studied a RS for movie lovers using neural
networks in collaborative filtering systems for consumers’ experiential decisions. The
experimental results reveal that it not only improves the accuracy of predicting movie ratings but
also increases data transfer rates and provides richer user experiences.
 6. EXPERIMENT AND RESULT

We apply decision tree algorithm with ten different models of dividing the dataset to training
and testing Ranging from (90:10) to (10:90) and the result as shown in table2.

Model Results for 5 attributes and the class Results for 12 attributes and the class
Num. Len. Mean Num. Len. Mean
Model Data
accuracy of of Square accuracy of of Square
num. allocation
rules rules Error rules rules Error
1- 90:10 66 % 37 19 0.4715 65.7895 % 63 32 0.5064
2- 80:20 67.6768 % 37 19 0.4903 58.6667 % 63 32 0.5103
3- 70:30 69.1275 % 37 19 0.4762 52.2124 % 63 32 0.5816
4- 60:40 59.0909 % 37 19 0.4968 54.6667 % 63 32 0.5679
5- 50:50 63.7097 % 37 19 0.489 61.1702 % 63 32 0.5181
6- 40:60 64.094 % 37 19 0.4795 55.3097 % 63 32 0.6245
7- 30:70 57.6369 % 37 19 0.502 55.5133 % 63 32 0.5606
8- 20:80 60.7053 % 37 19 0.5181 52.4917 % 63 32 0.563
9- 10:90 54.7085 % 37 19 0.5498 51.4793 % 63 32 0.6231
10- 66:34 65.6805 % 37 19 0.4725 53.125 % 63 32 0.5394
Table 2. DT accuracy Results for 10 models using 5 attribute and 12 attribute

Then apply Naïve Bayes algorithm with ten different attempts of dividing the dataset to training
and testing Ranging from (90:10) to (10:90) and the result as shown in table3.

Results for 5 attributes and the Results for 12 attributes and

Model
class the class
MSE(Mean
Model Data MSE(Mean
accuracy accuracy Square
num. allocation Square Error)
Error)
1- 90:10 72 % 0.4354 57.8947 % 0.5032
2- 80:20 65.6566 % 0.4593 62.6667 % 0.475
3- 70:30 64.4295 % 0.461 53.0973 % 0.5034
4- 60:40 62.6263 % 0.4832 56 % 0.4998
5- 50:50 61.2903 % 0.482 58.5106 % 0.4964
6- 40:60 63.4228 % 0.4799 58.8496 % 0.4987
7- 30:70 61.9597 % 0.4832 55.8935 % 0.5158
8- 20:80 60.2015 % 0.5065 53.4884 % 0.5381
9- 10:90 56.9507 % 0.5331 51.4793 % 0.5698
10- 66:34 65.6805 % 0.4601 52.3438 % 0.5097
Table 3. NB accuracy Results for 10 models using 5 attribute and 12 attribute
Lastly, we apply Neural Network (Multi-Layer Perceptron) algorithm using two Hidden layers
with ten different attempts of dividing the dataset to training and testing Ranging from (90:10) to
(10:90) and the result as shown in table4.

Model Results for 12

Results for 5 attributes and the class
attributes and the class
Model Data Num. Activation Learning Accuracy MSE Accuracy MSE
num. allocation Hidden Function rate (Mean (Mean
Nodes Square Square
Error) Error)
1- 90:10 2 Sigmoid 0.3 76 % 0.4316 44.7368 % 0.5531
2- 80:20 2 Sigmoid 0.3 66.6667 % 0.5075 56 % 0.5111
3- 70:30 2 Sigmoid 0.3 65.1007 % 0.498 46.0177 % 0.5603
4- 60:40 2 Sigmoid 0.3 51.5152 % 0.5185 57.3333 % 0.539
5- 50:50 2 Sigmoid 0.3 58.4677 % 0.5088 53.7234 % 0.6002
6- 40:60 2 Sigmoid 0.3 52.349 % 0.527 50 % 0.6431
7- 30:70 2 Sigmoid 0.3 53.6023 % 0.5416 52.4715 % 0.6027
8- 20:80 2 Sigmoid 0.3 54.1562 % 0.562 50.1661 % 0.658
9- 10:90 2 Sigmoid 0.3 57.6233 % 0.5735 52.3669 % 0.6466
10- 66:34 2 Sigmoid 0.3 64.497 % 0.4829 53.9063 % 0.5778
Table 4. ANN accuracy Results for 10 models using 5 attribute and 12 attribute
 Model Decision Tree Naïve Bayes Neural Network
90:10 66 % 72 % 76 %
80:20 67.6768 % 65.6566 % 66.6667 %
70:30 69.1275 % 64.4295 % 65.1007 %
60:40 59.0909 % 62.6263 % 51.5152 %
50:50 63.7097 % 61.2903 % 58.4677 %
40:60 64.094 % 63.4228 % 52.349 %
30:70 57.6369 % 61.9597 % 53.6023 %
20:80 60.7053 % 60.2015 % 54.1562 %
10:90 54.7085 % 56.9507 % 57.6233 %
66:34 65.6805 % 65.6805 % 64.497 %
Table 5 comparative of accuracy results for DT, NB and NN

80%
77%
74%
71%
68%
65% Decision Tree
62% Neural Network

59% Naïve Bayes

56%
53%
50%
90 80 70 60 50 40 30 20 10 66
Model Model Model Model Model Model Model Model Model Model
1 2 3 4 5 6 7 8 9 10

Figure 17. Comparative of accuracy results for DT, NB and NN

7 Conclusions
We used a dataset (dresses recommendation) from UCI, firstly we preprocess the dataset using
Weka then we put the resulting preprocessed data in classification models for three algorithms
(DT, NB, ANN). Because the Weka choose just (5) attribute by applying AttributeSelecton, We
try to put the dataset with original number of attributes (12) in classifiers models to see the
difference between the result using AttributeSelection and without using it, and as we can see
from results the dataset give better results for all methods after attribute selection, that mean the
existence of irrelevant attribute decrease the accuracy. Then we examine the best performance
for the three techniques by comparative the accuracy and mean square error using the dataset
after applying AttributeSelection. And as we can see our learning procedure is performing better
with Naïve Bayes model in terms of higher accuracy rate and lower mean square error rate.
References:

1. N Padhy, Dr. P. Mishra, “The Survey of Data Mining Applications And Feature Scope” , (IJCSEIT) International
Journal of Computer Science, Engineering and Information Technology, Vol.2, No.3, June 2012 .
2. Guerra L, McGarry M, Robles V, Bielza C, Larrañaga P, Yuste R. ,”Comparison between supervised and
unsupervised classifications of neuronal cell types: A case study. Developmental neurobiology” , 71(1): 71-
82,(2011).
3. Asmaa S. Hussein, Wail M. Omar, Xue Li, ModafarAti,” Efficient Chronic Disease Diagnosis Prediction and
Recommendation System” , IEEE EMBS International Conference on Biomedical Engineering and Sciences I
Langkawi I 17th - 19th December 2012.
4. Binge Cui, Xin Chen,” An Online Book Recommendation System Based on Web Service”, Sixth International
Conference on Fuzzy Systems and Knowledge Discovery,IEEE,2009.
5. Shahab Saquib Sohail, Jamshed Siddiqui, Rashid Ali,” Book Recommendation System Using OpinionMining
Technique”,IEEE,2013.
6. Sajal Halder, A. M. Jehad Sarkar, Young-Koo Lee,” Movie Recommendation System Based on Movie Swarm”,
Second International Conference on Cloud and Green Computing,IEEE,2012.
7. Abinash, Vineet, An Approach to Content Based Recommender Systems using Decision List based Classification
with k-DNF Rule Set, International Conference on Information Technology , 2014.
8. Nadav Golbandi, Yehuda Koren, Ronny Lempel,” Adaptive Bootstrapping of Recommender Systems Using
Decision Trees”, ACM 978-1-4503-0493-1/11/02,2011.
9. Iván Cantador, Desmond Elliott, Joemon M. Jose,” A Case Study of Exploiting Decision Trees for an Industrial
Recommender System”, Lilybank Gardens, Glasgow, G12 8QQ, UK,2009.
10. Sofia Visa, Anca Ralescu, Mircea Ionescu,” Investigating Learning Methods for Binary Data”, IEEE,2007.

11. Anika Gupta, Dr. Deepak Garg,” Applying Data Mining Techniques in Job Recommender System for Considering
Candidate Job Preferences”,IEEE,2014.
12. Mustansar Ali Ghazanfar and Adam Pr¨ugel-Bennett,” An Improved Switching Hybrid Recommender System Using
Naive Bayes Classifier and Collaborative Filtering”, School of Electronics and Computer Science, University of
Southampton, Highfield Campus, SO17 1BJ, United Kingdom,2010.
13. Sutheera Puntheeranurak, Pongpan Pitakpaisarnsin,” Time-aware Recommender System Using Naïve Bayes
Classifier Weighting Technique”, 2nd International Symposium on Computer, Communication, Control and
Automation (3CA 2013).
14. Meghna Khatri,” A Survey of Naïve Bayesian Algorithms for Similarity in Recommendation Systems”,
International Journal of Advanced Research in Computer Science and Software Engineering, Volume 2, Issue 5,
May 2012.
15. Anand Shanker Tewari, Tasif Sultan Ansari, Asim Gopal Barman,” Opinion Based Book Recommendation Using
Naïve Bayes Classifier”,IEEE,2014.
16. Sutheera Puntheeranurak, Supitchaya Sanprasert,” Hybrid Naive Bayes Classifier Weighting and Singular Value
Decomposition Technique for Recommender System”,IEEE,2011.
17. Maria Nadia Postorino, Giuseppe M. L. Sarne ,” A Neural Network Hybrid Recommender System, Proceedings of
the 2011 conference on Neural Nets WIRN10: Proceedings of the 20th Italian Workshop on Neural Net,2012.
18. Anant Gupta, Dr. B. K. Tripathy,” A Generic Hybrid Recommender System based onNeural Networks”,IEEE.2014.
19. M.K.Kavitha Devi, R.Thirumalai Samy, S.Vinoth Kumar, Dr.P.Venkatesh,” Probabilistic Neural Network approach
to Alleviate sparsity and cold start problems in Collaborative Recommender Systems”,IEEE,2010.
20. Alberto Y. Hata, Danilo Habermann, Fernando S. Osorio1, and Denis F. Wolf,” Road Geometry Classification using
ANN”, 20141EEE Intelligent Vehicles Symposium (IV)June 8-11, 2014. Dearborn, Michigan, USA.
21. Arthur J. Lina, Chien-Lung Hsub, Eldon Y. Lic,,” Improving the effectiveness of experiential decisions by
recommendation systems”,ACM,2014.
22. Machine Learning Repository (UCI) , https://archive.ics.uci.edu/ml/index.html.
23. R. Kirkby,E. Frank,” WEKA Explorer User”, University of Waikato, November 9, 2004.
Appendix

Technique Mean Square Error

Decision Tree 0.49457
Naïve Bayes 0.47837
Neural Network 0.51514
Table 6. Comparative of Mean Accuracy for DT, NB and NN

Mean Accuracy

80.00%
75.00%
70.00%
65.00%
60.00%
55.00%
50.00%
45.00% Mean Accuracy
40.00%
35.00%
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
Decision Tree Naïve Bayes Neural Network

Figure 18. Comparative of Mean for Accuracy and Mean Square Error for DT, NB and NN
 Figure 19. Example of raw dataset

1.2

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25
-0.2

Figure 20. Sample of raw dataset

 Figure 21. Example of transformation data from nominal to numeric

1.2

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25
-0.2

Figure 22. Sample of transformed dataset

 Figure 25. Example of dataset after discretization and attributes selection

1.2

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25 30 35 40
-0.2

Figure 26. Sample of dataset after discretization and attribute selection

View publication stats