Intelligent and Converged Networks ISSN 2708-6240

2023, 4(4): 296−304 DOI: 10.23919/ICN.2023.0024

Design and analysis of a recommendation system based on

collaborative filtering techniques for big data
Najia Khouibiri, Yousef Farhaoui*, and Ahmad El Allaoui

Abstract: Online search has become very popular, and users can easily search for any movie title; however, to easily
search for moving titles, users have to select a title that suits their taste. Otherwise, people will have difficulty choosing
the film they want to watch. The process of choosing or searching for a film in a large film database is currently time-
consuming and tedious. Users spend extensive time on the internet or on several movie viewing sites without success
until they find a film that matches their taste. This happens especially because humans are confused about choosing
things and quickly change their minds. Hence, the recommendation system becomes critical. This study aims to reduce
user effort and facilitate the movie research task. Further, we used the root mean square error scale to evaluate and
compare different models adopted in this paper. These models were employed with the aim of developing a
classification model for predicting movies. Thus, we tested and evaluated several cooperative filtering techniques. We
used four approaches to implement sparse matrix completion algorithms: k- nearest neighbors, matrix factorization,
co-clustering, and slope-one.

Key words: recommendation system; machine learning; collaborative filtering (CF); decision support system; big data

1 Introduction confusing[1].
In this study, we present a film recommendation
With the advent of big data and technological
system based on Collaborative Filtering (CF)
developments that marked the end of the 20th century
techniques. To this end, we implemented, tested, and
and the beginning of this century, the amount of data to
be exploited or analyzed has become very voluminous. evaluated several machine learning algorithms to
Knowing what data to look for and where to find them develop a predictive film provider rating model. The
is usually tedious. One such data searching process remainder of this paper is organized as follows. A
includes selecting or searching for an online film from literature review of movie recommendation systems is
a large film database, which makes users spend long provided in Section 2. In Section 3, We present the
hours on the internet or on many movie viewing sites methodology that is employed, along with a discussion
without success until they find a film that suits their on machine learning models and two evaluation
taste. Therefore, film recommendation systems aim to metrics. Section 4 discusses the results obtained in this
assist film lovers by suggesting which movie to watch study. Finally, Section 5 presents the conclusions and
without going through the lengthy film selection future studies[2].
process from a huge series of movies that extends to
2 Research background
thousands and millions, which is time-consuming and
Najia Khouibiri, Yousef Farhaoui, and Ahmad El Allaoui are 2.1 Related work
with STI Laboratory, IDM, T-IDMS, Faculty of Sciences and
Techniques Errachidia, Moulay Ismail University, Meknes Several studies have been conducted to recommend
5003, Morocco. E-mail: [email protected]; films.
[email protected]; [email protected].
For example, Ref. [3] suggests a movie
* To whom correspondence should be addressed.
Manuscript received: 2023-04-25; accepted: 2023-06-21 recommendation system that predicts the user

© All articles included in the journal are copyrighted to the ITU and TUP. This work is available under the CC BY-NC-ND 3.0 IGO license:
https://creativecommons.org/licenses/by-nc-nd/3.0/igo/.
 Najia Khouibiri et al.: Design and analysis of a recommendation system based on collaborative filtering techniques... 297

preference for a film based on different parameters that places greater emphasis on the analysis of
using the K- means clustering and k- nearest neighbor elements that contribute to generating predictions. For
(KNN) algorithms. the CB technique, the recommendation is based on the
A hybrid recommendation system proposed in Ref. user’s profile using features extracted from the content
[4] is built by combining two techniques, CF and of items that the user has evaluated in the past[11, 12].
content-based filtering (CB), to provide accurate Subsequently, it builds a user interest profile (see
recommendations for movies. The content filtering part Fig. 1).
of the system has been adopted to train neural networks 2.2.2 Collaborative filtering
representing individual user preferences. Filtering CF is an approach based on the sharing of opinions
results were combined using Boolean and fuzzy among users. It follows the principle of “ word of
aggregation operators. The data adopted in this model mouth” that people always practice to build an opinion
led to highly accurate predictions. on a product or service they do not know. The basic
Another study constructs a recommendation system premise of this method is that another user’s viewpoint
based on cosine similarity using KNN with the support can be used to provide a reasonable forecast of
of CF technique simultaneously to eliminate the preferences to an active user for an item that they have
disadvantages of CB filtering[5]. Some scholars
not yet evaluated. This method assumes that if users
suggested the development of a recommendation
have the same preferences for a set of items, they will
system based on multiple algorithms to obtain
probably have the same preferences for another set of
groupings, such as K- means, mini-batch K-means,
items that they have not evaluated yet[13, 14] . For
birch, affinity propagation, and other algorithms[6].
example, imagine that Ahmed’s neighbors discover
Additionally, several approaches have been presented
that a newly opened restaurant in their neighborhood is
to improve K- means so that not every cluster can
a success; he will decide to try it. However, if most of
dramatically augment the variance. For movies, this
his neighbors consider it a failure, he may decide not to
system is restricted to the use of groups based on type
and tags. go there. Similarly, CF techniques recommend items to
Most of the above studies employ CF approaches, the current user that are appreciated by users with the
such as matrix factorization neighborhood-based same tastes (see Fig. 2).
algorithms[7]. Other methods can be employed to 2.2.3 Hybrid recommendation system
predict missing viewer evaluations and find the list of A hybrid recommendation system combines two or
movies that the user would like to watch. The main more different referral approaches (CF and CB). The
contribution of our study is the testing and evaluation earlier approaches had various drawbacks, such as cold
of several strategies, including co-clustering and slope- start or data scarcity. These issues are frequently
one methods[8]. resolved by combining two or more techniques.
Moreover, with this hybridization, it is feasible to
2.2 Recommendation system
The recommendation system is a valuable tool that
Search, like, consult, …
provides the user with a list of suggestions and directs
them to a group of sources that may be useful and
Client
interesting to them, which can be difficult to reach in a Similar articles
short period of time within the big data space. For this
purpose, one of the following methods is used: CB, CF, Recommended
or hybrid approaches[9, 10].
2.2.1 Content-based filtering
The CB technique is a domain-dependent algorithm Fig. 1 Content-based recommendation system.
 298 Intelligent and Converged Networks, 2023, 4(4): 296−304

3.2 Description of the dataset

Similar users
The MovieLens† dataset contains data from the
Buy Buy
MovieLens platform collected by the GroupLens
Research Laboratory§, a human–computer interaction
Recommended research laboratory established by the University of
Minnesota. Its main purpose includes collecting data to
Protein
study recommendation systems. The GroupLens
Research Laboratory is considered one of the first
Fig. 2 Collaborative recommendation system.
laboratories to study automated recommendation
improve the recommendations’ quality[15–17]. systems[25, 26] . The MovieLens platform collects user
voices, which are then shared through several datasets.
3 Methodology The dataset displayed varies in terms of volume and
update date[27]. For this study, we visited the
3.1 Data selection
GroupLens website and selected one of the recent and
To create recommendations, it is necessary first to recommended data¶ for this research (released on
search for explicit data that can be worked on. For this 12/2019), where 25 million ratings and 1 million tag
purpose and for our choice to be based on a careful applications were applied to 62 000 films by 162 000
study, we rely on the results presented in a previous users.
study[18–20], where the most common and most used 3.3 Basic preparation and data exploration
dataset was studied by the authors in study
The preparation step is a key initial phase to prepare
recommendation systems[21–23].
the dataset for processing and in-depth analysis. Let us
As shown in Fig. 3, MovieLens and Amazon datasets
import our dataset files first because we will do the
are the most popular among researchers and data
same[28].
scientists for conducting diverse experiments, reaching
Now, we have two text files: movies.csv and
popularities of 40% and 35%, respectively. Because we ratings.csv. We perform a simple exploration of our
are concerned with movie recommendations and not data to know its content, understand its structure and
with users’ opinions about products, it is obvious that draw the required statistics for easier interpretation of
our choice will fall in the MovieLens dataset, which is the results.
the dataset that we adopt throughout this study[24, 25]. As there are no missing values, the data cleaning
Popularity of recommendation systems datasets process will not be performed for this dataset because it
ACM RecSys Conference 2017 & 2018; Full and short papers is as clean as possible.
Our movies.csv file contains the table of films, which
n=88 papers using on or more dataset(s)
Amount of papers
using the dataset

includes three columns: the film identifier (movieId),

40%
35% title, genre, and the title’s derived release date. It also
contains 62 423 movies.
13% 11% We do not know the movie selection criteria for the
8% 6% 5% 5% 5% MovieLens dataset, but it seems that most of the films
are famous, and others are old, dating back to 1903.
lp
ar n

ar !
a s

FM
ns
ry

or
ll V oo

g
ll V zo
ll V en

Ye
.)

.)
A r.)

in
ta

vis

io
(A ah

st
(A ma
(A ieL

ss

Additionally, the sorting of films was based on the date

rie

in

La
ad

ro

Ep
op
ov

ip

-c
M

Pr

of addition, not the date of release.

Tr

ok
Bo

Dataset
The second file, ratings.csv, contains a table of user
† http://www.movielens.org/
Fig. 3 Most common datasets for studying § http://www.grouplens.org/
recommendation systems. ¶ https://www.grouplens.org/datasets/movielens/25m/
 Najia Khouibiri et al.: Design and analysis of a recommendation system based on collaborative filtering techniques... 299

ratings for movies. It has four columns and 697 561 developing a movie recommendation system based on
rows. The columns are movieId, rating (users can rate the movie rating model.
movies from 0.5 to 5), timestamp (the time with the (1) Gather data in the form of explicit movie viewer
date of voting), and userId. Then, we removed the ratings (user ratings) and then prepare and explore it
timestamps column because it serves no purpose for us. beforehand.
It can be seen that this dataset contains movieId, the (2) Test and evaluate various machine learning
title of the film and its genre. We need a dataset models on ready data using a cross-validation
containing the userId (to extract user data; thus, we will technique and then choose the model with the best
be able to use user data to increase the precision of performance.
recommendations because MovieLens does not offer a (3) To develop the desired recommendation system,
table relating to users), movie titles, and notes. This we ultimately deployed the trained model. Figure 4
information is included in two different data frame shows the various phases of the suggested solution.
objects: df_ratings and df_movies. To obtain the 3.4.1 Machine learning models
desired information in a single data frame (Table 1), Systems using the CF technique should compare
we can merge these two data frame objects on the objects that are significantly different from one
movieId column, as it is common for these data frames. another: items in relation to users. The neighborhood
We can do this using the merge() function of the method and latent factor models are the two main
Pandas library. strategies for facilitating such a comparative
In our study, finding the best machine learning evaluation. Additionally, the co-clustering and slope-
model that can accurately predict the missing ratings is one methods have been suggested in the literature to
a difficult task. This is the reason we remove users with deal with the recommendation issue. The machine
only one review (only retain viewers who have more learning models that we have employed to forecast
reviews than the average number of reviews per missing ratings have been presented in this part.
viewer). However, there is nearly 99% sparsity in the Neighbors-based models. There are two main stages
built movies rating matrix. for suggesting recommendations based on the
neighbor’s model. The first stage is to establish the
3.4 Suggested solution
neighborhood, and the second stage is to make
In a study on the health care provider’s recommendations.
recommendation system[28], several methods are During the neighborhood build process, similarity
studied, e.g., the neighborhood method and latent between users (called a user-based approach) or
factor models. Additionally, the proposed solution is elements (item-based approach) is measured. The two
applied in four stages. In our study, we worked on most widely used similarity measures are Pearson’s
larger data (including the recommendation system to correlation (PC) coefficient (Eq. (1)) and cosine-based
address the problem of user loss amid big data) and similarity (Eq. (2)).
tried to work on some algorithms that are applied in the ∑
n
aforementioned study and test them on a large (xi − x′ )(yi − y′ )
i=1
dataset and other data in a movie recommendation PC (x, y) = v
t v
t (1)
∑
n ∑
n
system[29, 30]. The following strategy is suggested for (xi − x ) ′ 2 ′ 2
(yi − y )
i=1 i=1
Table 1 Statistic results on the final dataset.
Entry Number where x and y are two n-pointed vectors. The average
Movie 62 000 values of vectors x and y are represented by x' and y',
Unique user 100 000 respectively. PC determines the relationship between
Rating 400 000
two sets of data, x and y.
 300 Intelligent and Converged Networks, 2023, 4(4): 296−304

Machine
learning model

1 Neighbors-based
model
Preparation & Cross validation Co-clustering-
exploration
based model

Dataset Ready dataset Latent factors

model

Slope-one based
model
Evaluation

Construction

Our movie
3
recommendation system
Best model

Fig. 4 Diagram to clarify the suggested solution.

∑
n
Ei × Fi Latent factor models. The goal of latent factor
cos (θ) = v
i=1
v (2) models, which are an alternative method to CF, is to
t t
∑
n ∑
n
identify latent (hidden) characteristics of the data.
2 2
Ei Fi
i=1 i=1 This provides explicit explanations about data
showing what users are feeling about an element,
where E and F are two groups of n data points or n
which is typically stored in ratings matrix form. The
characteristic values. Ei and Fi represent the values of
feature i in sets E and F, respectively. The next phase is matrix factor is among the most popular methods used
to predict user u’ s evaluation r̂ui that they will most for identifying latent factors. Figure 5 illustrates the
likely give to element i . Among the methods that can principle of the matrix factorization method. In this
be used is the use of calculated similarities and study, we limited ourselves to testing only one of the
corresponding evaluations. Many differences are matrix factorization methods, i.e., non-negative matrix
possible by including biases like means, Z-score, or the factorization (NMF).
median user/item evaluations. In this study, we only By calculating the point product of two vectors
tested one difference, which is the KNN baseline. related to qi and pu, It is straightforward to calculate the
The predicted rating is determined using Eq. (3) if user’s prediction of their evaluation for an element i, as
the methodology is user-based; however, if the indicated in the formula below.
methodology is item-based, the predicted rating is r̂ui = pTu qi ,
determined using Eq. (4).
∑ where qi is a vector related to element i , and pu is a
sim (u, v) · rvi
vector related to user u.
v∈Nik (u)
r̂ui = ∑ (3) Slope-one based model. Slope-one predictors are
sim (u, v)
suggested to use for collaborative rating-based filtering
v∈Nik (u)
∑ algorithms to reduce the fitting problem, increase
sim (i, j) · ru j efficiency, and facilitate and implement
j∈Nuk (i) recommendation systems. Based on the use of a simple
r̂ui = ∑ (4)
sim (i, j) form of regression, they are regarded as simple
j∈Nuk (i) approaches to implementing a prediction. The median
 Najia Khouibiri et al.: Design and analysis of a recommendation system based on collaborative filtering techniques... 301

Item
A1 A2 A3 A4 A1 A2 A3 A4

B1 4.5 2.0 B1 1.2 0.8 B1 1.5 1.2 1.0 0.8

B2 4.0 3.5 B2 1.4 0.9 B2 1.7 0.6 1.1 0.4

= ×
User

B3 5.0 2.0 B3 1.5 1.0

B4 3.5 4.0 1.0 B4 1.2 0.8

Rating matrix User matrix Item matrix

Fig. 5 Matrix factorization technique.

contrast between the degrees of the two elements is the evaluation, to which the user u is considered to belong,
only free parameter. In some cases, it has proved to be and Ci represents the average evaluation of the cluster,
considerably more precise than the linear regression of to which element i is considered to belong.
the degrees of one element to the degrees of another 3.4.2 Evaluation metric
element. The most popular and widely used scales for evaluating
Therefore, the prediction is calculated using the recommendation systems are the root mean squared
following relationship: error (RMSE) and mean absolute error scales. In this
∑ 1
r̂u,i = µu + dev(i, j) (5) study, we used the RMSE scale to evaluate our
|Ri (u)| j∈Ri (u) recommendation system, which can be calculated using
where Ri(u) is the collection of pertinent elements (i.e., Eq. (8).
√∑
the collection of elements j rated by u and shared with
(r̂ui − rui )2
at least one user i ), and dev(i, j ) represents the RMSE = (8)
n
difference of average rating between elements i and j,
where r̂ui is the expected user u rating for item i, rui is
and it is calculated using the following Eq. (6):
the rating that was actually given, and n is the volume
1 ∑
dev (i, j) = ru,i − ru, j (6) of the test set (size).
Ui j u∈Ui j In this paper, on all of our samples, we perform a 5-
where Uij represents all users that rated items i and j. cross-validation RMSE. We try to train our model
Co-clustering-based model. In the field of data using 80% of the data, and the remaining 20% is used
mining, the term “clustering” denotes the process of for testing the accuracy.
grouping objects into similar objects belonging to the
4 Result
same group or cluster. Clustering is an unsupervised
learning technique. According to the type of data, The outcomes of our testing and evaluation of various
different aggregation techniques could be applied. The methods of the primary models mentioned above are
user element rating matrix is used as data in the case of summarized in Table 2.
CF. Users and elements are determined by certain Cu,i It can be seen that the baseline user-based CF KNN
co-clusters, Ci clusters, and certain Cu using a bi- is the best model in terms of RMSE value. Our aim is
clustering technique. Clusters are selected using an to find the best (optimal) metrics for each of the
uncomplicated optimization technique, similar to K- models. A detailed summary of the findings is provided
means[17]. We can calculate predictive rating using the in Table 2.
following Eq. (7):
5 Conclusion and perspective
( ) ( )
r̂ui = Cui + µu − Cu + µi − Ci (7)
This study aimed to develop recommendation systems
where Cui represents the median evaluation for the Cui using the CF approach with several machine learning
co-cluster, Cu represents the cluster’s median models. Our testing experiments proved that the
 302 Intelligent and Converged Networks, 2023, 4(4): 296−304

Table 2 Results summary.

Latent factors
Metric Neighbors-based model Co-clustering based model Slope-one based model
based model
Tested User-based CF KNN + Co-clustering based on Basic slope-one
NMF
approach baseline K-means algorithm algorithm
RMSE 0.8535 0.9131 0.8746 0.8864

models based on neighbors and latent factors have References

succeeded in providing more accurate [1] F. Furtado and A. Singh, Movie recommendation system
recommendations (i.e., with a low error rate). Other using machine learning, Int. J. Res. Ind. Eng., vol. 9, no. 1,
methods can also be adopted, such as slope-one and co- pp. 84–98.
clustering, to solve the problem of anticipated missing [2] A. Hessane, A. El Youssefi, Y. Farhaoui, B. Aghoutane,
ratings in the rating matrix for films by users. N. A. Ali, and A. Malik, Healthcare providers
recommender system based on collaborative filtering
However, our proposed recommendation system
techniques, in Machine Learning and Deep Learning in
suffers from several obstacles, including the method of
Medical Data Analytics and Healthcare Applications, O.
calculating similarities and cold start obstacles, which P. Jena, B. Bhushan, and U. Kose, Eds. Boca Raton, FL,
can only be solved by adopting more than one machine USA: CRC Press, 2022. pp. 261–274.
learning technique (there is a need to create a hybrid [3] R. Ahuja, A. Solanki, and A. Nayyar, Movie recommender
system) or relying on written surveys of people. This is system using K-means clustering AND K-nearest
done by merging survey programs with movie neighbor, in Proc. 2019 9th Int. Conf. Cloud Computing,
Data Science & Engineering (Confluence), Noida, India,
recommendation programs.
2019, pp. 263–268.
The task of recommending movies has been [4] C. Christakou, S. Vrettos, and A. Stafylopatis, A hybrid
daunting, and it will become more challenging in the movie recommender system based on neural networks, Int.
years to come because of the alarming increase in the J. Artif. Intell. Tools, vol. 16, no. 5, pp. 771–792, 2007.
volume of data. This means that the basis for [5] M. Gupta, A. Thakkar, Aashish, V. Gupta, and D. P. S.
recommending movies should not be limited only to Rathore, Movie recommender system using collaborative
the opinions of similar users, but more information, filtering, in Proc. 2020 Int. Conf. Electronics and
Sustainable Communication Systems (ICESC),
such as age and gender, should be considered. Why not
Coimbatore, India, 2020, pp. 415–420.
also consider the health status of the user? This may [6] D. Cintia Ganesha Putri, J.-S. Leu, and P. Seda, Design of
sound somewhat strange, but it is realistic. How many an unsupervised machine learning-based movie
users have died just because they watched a comedy or recommender system, Symmetry, vol. 12, no. 2, p. 185,
horror movie recommended to them by a similar friend, 2020.
without taking into account that they are asthmatic or [7] F. O. Isinkaye, Y. O. Folajimi, and B. A. Ojokoh,

heart patients? Moreover, how many teenagers, due to Recommendation systems: Principles, methods and
evaluation, Egypt. Inform. J., vol. 16, no. 3, pp. 261–273,
the error of recommending films that do not agree with
2015.
their age, have committed suicide? [8] I. Benouaret, Un système de recommandation contextuel
In essence, the more information we collect, the et composite pour la visite personnalisée de sites culturels,
greater the significance of similarity calculations, the (in French), Ph. D. dissertation, University of Technology
recommendation system is more accurate and safer for of Compiègne, France, 2017, pp. 181.
the user’s life because human life does not accept any [9] M. Baidada, K. Mansouri, and F. Poirier, Hybrid filtering

room for error. Therefore, relying solely on machine recommendation system in an educational context, Int. J.
Web Based Learn. Teach. Technol., vol. 17, no. 1, pp.
learning techniques is insufficient. Instead, we must
1–17, 2022.
look forward to developing a hybrid recommendation [10] J. Beel and V. Brunel, Data pruning in recommender
system that adopts various deep learning techniques systems research: Best-practice or malpractice? in Proc.
and integrates data mining techniques to eliminate the ACM RecSys 2019 Late-Breaking Results & 13th ACM
cold start problem. Conf. Recommender Systems, Copenhagen, Denmark,
 Najia Khouibiri et al.: Design and analysis of a recommendation system based on collaborative filtering techniques... 303

2019, pp. 26–30. [20] Y. Farhaoui, S. Ojo, L. A. Akinyemi, and A. L. Imoize,

[11] J. Bastin, Etude des systèmes de recommandations et mise Editorial, Big Data Mining and Analytics, vol. 6, no. 3, pp.
en pratique des algorithms, (in French), master i–ii, 2023.
dissertation, University of Liège, Belgium, 2020. [21] Y. Farhaoui, Intrusion prevention system inspired immune
[12] Y. Koren, Factorization meets the neighborhood: a systems, Indones. J. Electr. Eng. Comput. Sci., vol. 2, no.
multifaceted collaborative filtering model, in Proc. 14th 1, p. 168, 2016.
ACM SIGKDD Int. Conf. Knowledge discovery and data [22] Y. Farhaoui, Big data analytics applied for control
mining, Las Vegas, NV, USA, 2008, pp. 426–434. systems, in Proc. Int. Conf. Advanced Information
[13] O. A. Montesinos-López, A. Montesinos-López, J. Crossa, Technology, Services and Systems (AIT2S-17), Tangier,
J. C. Montesinos-López, D. Mota-Sanchez, F. Estrada- Morocco, 2017, pp. 408–415.
González, J. Gillberg, R. Singh, S. Mondal, and P. Juliana, [23] Y. Farhaoui, B. Bhushan, M. Fattah, and B. Aghoutane,
Prediction of multiple-trait and multiple-environment Editorial, Big Data Mining and Analytics, vol. 5, no. 4, pp.
genomic data using recommender systems, i–ii, 2022.
G3 Bethesda Md, vol. 8, no. 1, pp. 131–147, 2018. [24] S. S. Alaoui, Y. Farhaoui, and B. Aksasse, Hate speech
[14] P. Wang and H. Ye, A personalized recommendation detection using text mining and machine learning, Int. J.
algorithm combining slope one scheme and user based Decis. Support. Syst. Technol., vol. 14, no. 1, pp. 1–20,
collaborative filtering, in Proc. 2009 Int. Conf. Industrial 2022.
and Information Systems, Haikou, China, 2009, pp. [25] S. S. Alaoui, Y. Farhaoui, and B. Aksasse, Data openness
152–154. for efficient e-governance in the age of big data, Int. J.
[15] D. Lemire and A. MacLachlan, Slope one predictors for Cloud Comput., vol. 10, nos. 5&6, p. 522, 2021.
online rating-based collaborative filtering, in Proc. 2005 [26] A. El Mouatasim and Y. Farhaoui, Nesterov step reduced
SIAM Int. Conf. Data Mining, Newport Beach, CA, USA, gradient algorithm for convex programming problems, in
2005, pp. 471–475. Proc. 3rd Int. Conf. Big Data and Networks Technologies
[16] M. Al-Ghamdi, H. Elazhary, and A. Mojahed, Evaluation (BDNT 2019), Leuven, Belgium, 2019. pp. 140–148.
of collaborative filtering for recommender systems, Int. J. [27] A. Tarik, and Y. Farhaoui, Recommender system for
Adv. Comput. Sci. Appl., vol. 12, no. 3, pp. 559–565, 2021. orientation student, in Proc. 3rd Int. Conf. Big Data and
[17] T. George and S. Merugu, A scalable collaborative Networks Technologies (BDNT 2019), Leuven, Belgium,
filtering framework based on co-clustering, in Proc. 5th 2019, pp. 367–370.
IEEE Int. Conf. Data Mining (ICDM'05), Houston, TX, [28] S. Sossi Alaoui, Y. Farhaoui, and B. Aksasse, A
USA, 2005. comparative study of the four well-known classification
[18] G. Shani and A. Gunawardana, Evaluating algorithms in data mining, in Proc. Int. Conf. Advanced
recommendation systems, in Recommender systems Information Technology, Services and Systems (AIT2S-
handbook, F. Ricci, L. Rokach, B. Shapira, and P. B. 17), Tangier, Morocco, 2017. pp. 362–373.
Kantor, Eds. Boston, MA, USA: Springer, 2011, pp. [29] Y. Farhaoui, Teaching computer sciences in Morocco: An
257–297. overview, IT Prof., vol. 19, no. 4, pp. 12–15, 2017.
[19] Y. Farhaoui, Design and implementation of an intrusion [30] Y. Farhaoui, Securing a local area network by IDPS open
prevention system, Int. J. Netw. Secur., vol. 19, no. 5, pp. source, Procedia Comput. Sci., vol. 110, pp. 416–421,
675–683, 2017. 2017.

Najia Khouibiri received the MEng Ahmad El Allaoui is now an assistant

degree in decision information systems and professor at the Department of Computer
imaging from Moulay Ismail University, Science, Faculty of Sciences and
Morocco in 2022. She is a PhD candidate Techniques Errachidia, Moulay Ismail
at the Engineering Science and University, Morocco. He is an IDMS
Technology Laboratory, IDMS Team, Team member. He focuses in semantic
Faculty of Sciences and Techniques image segmentation, medical imaging,
Errachidia, Moulay Ismail University, classification algorithms, segmentation,
Morocco. Her research interests include business intelligence, image processing, evolutionary algorithms, and genetic
cloud computing, and big data. algorithms.
 304 Intelligent and Converged Networks, 2023, 4(4): 296−304

Yousef Farhaoui obtained the PhD degree

in computer security from Ibn Zohr
University of Science, Morocco in 2012.
He is now a professor at Faculty of
Sciences and Techniques Errachidia,
Moulay Ismail University, Morocco. He is
the chair of IDMS Team and director of
STI Laboratory. He is the local publishing
and research coordinator, Cambridge International Academics
Ltd., UK. His research interests include learning, e-learning,
computer security, big data analytics, and business intelligence.
He has published 3 books on computer science. He is also a
member of the scientific committee of several international
congresses and a member of various international associations.
He has authored 7 books and many book chapters with reputed
publishers, such as Springer and IGI. He is a reviewer for IEEE,
IET, Springer, Inderscience, and Elsevier journals. He is also the
guest editor of many journals with Wiley, Springer, and
Inderscience. He has been the general chair, session chair, and
panelist in several conferences. He is a senior member of IEEE,
IET, ACM, and EAI Research Group.