IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 2, April 2025, pp. 1654~1662

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i2.pp1654-1662  1654

Artificial intelligence algorithms to predict customer

satisfaction: a comparative study

Othman Berrada Chakour1, Abdelaziz Ettaoufik2, Khalid Aissaoui1, Abderrahim Maizate1

1
RITM Laboratory/ESTC, CED ENSEM, Hassan II University, Casablanca, Morocco
2
LTIM Laboratory, Faculty of Sciences Ben M’sik, Hassan II University, Casablanca, Morocco

Article Info ABSTRACT

Article history: Customer satisfaction is the key for every business successful. Therefore,
keeping the current customer portfolio and expanding it over time is the
Received May 14, 2024 main goal for any business. Hence, we need first to satisfy these clients. The
Revised Nov 7, 2024 customer satisfaction helps to retain consumers of its products, increase the
Accepted Nov 14, 2024 life value of the customer, also make known its brand through positive word
of mouth to get a better reputation and thus increase turnover. For this
reason, several studies have been conducted on this subject to explore all
Keywords: tools and technologies that will help retain customers and reduce their churn
rate. Based on various customer satisfaction studies for different types of
Artificial intelligence businesses, this paper shows the review of promising research areas and
Churn prediction artificial intelligence (AI) application models in predicting customer
Customer satisfaction satisfaction. The results of this study allowed the identification of the best
Deep learning algorithms with the highest score of performance metrics that can be applied
Machine learning as part of the customer satisfaction prediction, through a detailed benchmark
performed. The result shows that random forest (RF) and gradient boost
(GB) algorithms in machine learning (ML) and convolutional neural
network - long short-term memory (CNN-LSTM) in deep learning (DL) are
giving the best performance. The most used metrics are accuracy and
F1-score. In addition, DL models outperform ML models in most cases.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Othman Berrada Chakour
RITM Laboratory/ESTC, CED ENSEM, Hassan II University
Casablanca, Morocco
Email: [email protected]

1. INTRODUCTION
Customer satisfaction is a term widely used in marketing industry, and represents a service or product
evaluation indicator, that needs to meet the client expectations. This assessment considers all the criteria that
matter to the customer and that he will experience once the product/service is delivered. Customer satisfaction
could have a positive influence on the retention rate and loyalty, also reduces churn rate and protects customers
from other competitors especially since the competition has never been so fierce, thanks to the internet. Client
satisfaction determines whether your business has a positive or negative brand image over the long term.
Furthermore, consumer expectations are constantly changing. Faced with the considerable and
limitless growth for data exchanged, customer specification in terms of responsiveness and efficiency became
denser. Faced with such a need for responsiveness and efficiency, it is obvious that the link with technologies
was quickly established. The previous researches results show that customer satisfaction prediction can be
analyzed using various types of machine learning (ML) and deep learning (DL) algorithms. Due to different
data formats (such as review data or survey data), different models are recommended for different application

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  1655

domains. Artificial intelligence (AI), ML and DL have become vital to remain competitive for financial
services companies around the globe.
The use of AI algorithms helps predict customer satisfaction and future behavior regarding the
purchase of a particular product again. Having a good customer satisfaction prediction model gives insight into
the future, which can help companies choose and adjust the strategies to implement and the right decisions to
make to achieve a high level of customer satisfaction. AI algorithms are used today in several sectors and
industries. In particular, the sub-domains of AI: ML and DL algorithms are used in banking sector to predict
customer churn [1]‒[3] and customer satisfaction [4], [5]. Therefore, they are used in the e-commerce industry
to predict the online shopper’s behavior to know if a customer will make a new purchase or not on the site
[6]‒[8]. Hence, ML and DL algorithms are applied in the telecommunication sector to predict customer churn
[9], in airlines domain to predict the satisfaction of plane passengers [10], [11], in transportation sector for
predicting car accidents [12], in agriculture field to predict demand and the products pricing [13], in healthcare
and smart home to detect falls [14], in smart manufacturing applications for predicting tool wear [15], in
entertainment domain for correcting postures of piano players [16], and several other fields.
Compared with other studies, our paper not only finds different relationships between factors
affecting customer satisfaction prediction, but also includes comparing the performance of various ML and
DL methods. This paper shows the comparative study of research domains and applications of AI models in
predicting customer satisfaction. This study began by defining the methods based on the prediction process.
In this section, we focused on the popular AI algorithms. In the second part of this article, we presented
various researches that uses ML and DL algorithms to predict customer satisfaction. Finally, we presented an
analysis of these researches through a benchmark that details for each article, the algorithms used, their types,
the date of publication of the article, the metrics evaluated and the optimal algorithm in terms of performance
based on the score of the different metrics.

2. METHOD
AI refers to attempts to build machines able to challenge the human on its intelligence. ML and DL
represent two particularly important steps in the evolution of AI. We define ML as a subset of AI developed
to imitate human intelligence. While DL is, a subset of ML based on artificial neural networks (ANNs).
Figure 1 represents the relationship between AI, ML and DL.
To predict customer satisfaction, it was necessary to study the different algorithms to define the
most efficient method. In this section, several popular ML algorithms such as logistic regression (LR),
decision tree (DT), random forest (RF), gradient boost (GB), k-nearest neighbors (KNN), support vector
machine (SVM), naive Bayes (NB), and DL algorithms such as ANN, convolutional neural network (CNN),
long short-term memory (LSTM), CNN-LSTM are discussed. Through this section, we will define the
previous mentioned algorithms and share examples of their application domain and results.

Figure 1. Relationship between AI, ML, and DL [17]

2.1. Machine learning algorithms

2.1.1. Logistic regression
LR is a ML algorithm based on mathematical model that estimates the probability of an occuring event
based on a set of explanatory variables. The result of the LR is a probability that could be between zero and one.
Thus, when the value is greater than 0.5, the event is likely going to occur but, if it is less than 0.5, it will not
[18]. The LR algorithm is used in several application domains. The best results are represented in the following
papers. LR was used to predict mobile operator’s customer churn rate in the telecommunication industry [9],
predict satisfaction of online banking customer in Bangladesh [4], evaluate the behavior of online shoppers in
the e-commerce domain [6] and analyze consumer sentiment using online feedback in the airlines industry [19].

2.1.2. Decision tree

DT is a ML algorithm that assist on decision-making. It gathers on the same graphic various options
related to a specific situation under a tree branches form, at each end of a branch, a potential decision. It
allows the distribution of a given population into homogeneous groups to achieve a goal without setting aside
Artificial intelligence algorithms to predict customer satisfaction: … (Othman Berrada Chakour)
1656  ISSN: 2252-8938

the discriminating characteristics of this population. It facilitates the read of a complex situation and it helps
to quickly execute the decision made [20]. DT is utilized in several application domains. The best
performances are showed in the subsequent researches. DT was used to assess the problem of predicting
credit card churn in banks [3], predict satisfaction of online banking customer in Bangladesh [4], predict
e-commerce customer satisfaction [7] and analyze the propensities of South Korea airlines customers for
predicting churn and customer satisfaction [16].

2.1.3. Random forest

RF is a prediction ML algorithm which was designed to obtain a reliable prediction via a system of
random subspaces. It consists of several DTs, independently trained on subsets of the training dataset. Each
DT produces an outcome, and it is the combination of the results that will give the final prediction. Each
model is randomly distributed into subsets of DTs [21]. The RF algorithm is employed in multiple
application scopes. The best results are represented in the following papers. RF was utilized to predict mobile
operator’s customer churn rate in the telecommunication industry [9], calculate the bank customer churn
prediction [1], analyze consumer sentiment using online feedback in the airlines industry [19], and evaluate
the behavior of customers for an European bank to predict churn rate [2].

2.1.4. Gradient boosting

GB is also a prediction ML algorithm. Unlike "random forest", it adjusts its training base gradually. It
uses the loss function gradient (or gradient descent) to correct the errors of the previous tree and thus optimize
the model results [22]. There are several types of GB algorithm. The most popular are: adaptive boosting
(AdaBoost) and eXtreme gradient boosting (XGBoost). The GB algorithms are utilized in several application
domains. The best results are presented in the following papers. XGBoost was employed to predict the churn
rate of customers in the telecommunications industry [9] and predict consumer repurchase behavior in the
e-commerce domain [23]. Moreover, AdaBoost was used to evaluate the behavior of customers of a European
bank to predict the churn rate [2] and assess the problem of predicting credit card churn in banks [3].

2.1.5. Support vector machine

SVM is a classification ML algorithm whose objective is to find the optimal hyperplane to separate
categories based on training data. It will focus on the closest points between the categories (vector supports).
This algorithm consists of finding the maximum margin between vector supports and draw the line in the
middle of it [24]. SVM is utilized in several application domains. The best performances are showed in the
subsequent researches. SVM was used to predict e-commerce customer satisfaction [7], evaluate bank
customer churn in the early stages by analyzing their behavior [25], predict satisfaction of online banking
customer in Bangladesh [4] and evaluate consumer sentiment using online feedback in the airlines industry [19].

2.1.6. K-nearest neighbors

KNN is also a classification ML algorithm that relies on the choice of the classification metric,
based only on training data. The principle of this model consists of choosing the nearest k data (Euclidean
distance) of the studied point to predict its value [26]. The KNN algorithm is employed in multiple
application scopes. The best results are represented in the following papers. KNN was utilized to assess the
problem of predicting credit card churn in banks [3], calculate the bank customer churn prediction [1], predict
the behavior of online shoppers in the e-commerce domain [6], and analyze the propensities of South Korea
airlines customers for predicting churn and customer satisfaction [16].

2.1.7. Naive Bayes

NB is a classification ML algorithm based on the Bayes theorem that determines the probability of an
event occurring from another event that has already occurred. Among the advantages of this algorithm is the ease
of understanding and implementation as it uses a very simple mathematical formula. Moreover, it is fast and does
not need a large dataset to output consistent results. However, this algorithm is called "naive" because it assumes
that there is no correlation between the data while this assumption is wrong in most cases [26]. The NB algorithm
is involved in various application domains. The best performances are showed in the subsequent researches. NB
was utilized to evaluate the problem of predicting churn for credit cards in banks [3], predict customer satisfaction
of internaltion airport of San Francisco in the airlines industry [15], analyze consumer sentiment using online
feedback in the airlines industry [19], and predict the behavior of online shoppers in the e-commerce domain [6].

2.2. Deep learning algorithms

2.2.1. Artificial neural network
ANN is a DL algorithm inspired by the structure of a human brain. It comes in the form of at least two
layers of neurons that, from a stream of learning data, will interact to learn to perform tasks [5]. The ANN

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1654-1662

Int J Artif Intell ISSN: 2252-8938  1657

algorithm is employed in different application domains. The best results are represented in the following papers.
The ANN was used to predict bank customer satisfaction [5], assess the problem of predicting credit card churn in
banks [3], predict customer churn based on behavioral model analysis in e-commerce domain [8], calculate the
bank customer churn prediction [1], and predict churn prediction of supermarket customer in retail industry [27].

2.2.2. Convolutional neural network

CNN is a DL algorithm that learns features from large amounts of data. CNN works by ingesting
and processing large amounts of data in a pattern and then extracting the most important features for
classification and detection. A typical CNN consists of three types of components: pooling layer,
convolutional layer, and fully connected layer. Each component has a different goal, performs a task based
on input data, and learns increasing amounts of complexity [28]. The CNN algorithm is utilized in various
application domains. The best performances are showed in the subsequent researches. CNN was involved to
predict customer satisfaction based on previous purchases [29], predict churn prediction of supermarket
customer in the retail industry [27], predict consumer repurchase behavior in the e-commerce domain [23] and
analyze the propensities of South Korea airlines customers for predicting churn and customer satisfaction [16].

2.2.3. Long short-term memory

LSTM is a DL algorithm which is a variation of recurrent neural networks (RNN). LSTM introduce
a new unit called a memory cell, which allows the network to store and access information over an extended
period. The memory cell of an LSTM consists of several gates: an input gate, an output gate, and a forget
gate. These gates regulate the flow of information within the memory cell, allowing control over what
information to remember and what to forget. This gives the LSTM the ability to memorize important
information over long sequences and ignore less relevant elements [30]. The LSTM algorithm is employed in
different application domains. The best results are represented in the following papers. LSTM was used to
predict consumer repurchase behavior in the e-commerce domain [23], analyze consumer sentiment using
online feedback in the airlines industry [19], predict user satisfaction based on sentiment analysis for social
media data in Saudi Arabia’s [31] and predict customer abandonment in the telecommunication industry [32].

2.2.4. Convolutional neural network - long short-term memory

CNN-LSTM is a DL algorithm that represents a sequence of two networks: CNN and LSTM. This
type of data has a memory mechanism that allows it to retain information from one prediction to another.
Therefore, CNN-LSTM is a CNN whose final output is sent directly into an LSTM, which will then realize its
prediction [33]. The CNN-LSTM algorithm is employed in various application domains. The best performances
are showed in the subsequent researches. CNN-LSTM was used to analyze the propensities of South Korea
airlines customers for predicting churn and customer satisfaction [16], analyze consumer sentiment using online
feedback in the airlines industry [19], predict user Satisfaction based on sentiment analysis for social media data
in Saudi Arabia’s [31], and predict consumer repurchase behavior in the e-commerce domain [23].

2.3. Performance evaluation of the model

Evaluating the performance of a ML model is one of the important steps in building an effective ML
model. Various metrics are used to assess the quality of the model. These performance metrics help us
understand how well our model performs for the given data. There are a variety of evaluation metrics
available in ML. These metrics are based on 4 concepts: true positive (TP) denotes that the customer is
satisfied, and the model predicts that the customer is satisfied; true negative (TN) indicates that the customer
is not satisfied, and the model predicts that the customer is not satisfied; false positive (FP) shows that the
customer is not satisfied, but the model predicts that the customer is satisfied; and false negative (FN)
denotes that the customer is satisfied, but the model predicts that the client is not satisfied. TP, TN, FP, and
FN are shown in Figure 2 that represents the confusion matrix.

2.3.1. Accuracy
Accuracy is an important and intuitive metric as it calculates the proportion of correct predictions
out of the total number of predictions [34]. It measures how often a ML model correctly predicts the
outcome. This metric is simple to calculate and understand. Accuracy can be calculated by (1).
∑ 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒+∑ 𝑇𝑟𝑢𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = × 100 (1)
∑ 𝑇𝑜𝑡𝑎𝑙 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛

2.3.2. Precision
Precision appears to be an important metric in monitoring model performance, as it generally
calculates the ratio between the best predicted results and the effective results [34]. It measures the frequency
of ML predicting the positive cases. Precision can be calculated by (2).

Artificial intelligence algorithms to predict customer satisfaction: … (Othman Berrada Chakour)

1658  ISSN: 2252-8938

∑ 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = × 100 (2)
∑ 𝐹𝑎𝑙𝑠𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠+ ∑ 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒

2.3.3. Recall
Recall is the ratio of the number of correctly predicted positive instances (TP) to the total number of
actual positive instances [34]. In other words, it measures the completeness of the ML model when it
identifies positive cases. Recall can be calculated by (3).
∑ 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒
𝑅𝑒𝑐𝑎𝑙𝑙 = × 100 (3)
∑ 𝑇𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠+ ∑ 𝐹𝑎𝑙𝑠𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒

2.3.4. F1 score
F1 score provides a single metric that encapsulates both dimensions of a model’s accuracy. It is a
weighted comparison of average precision and recall [34]. This metric ensures that precision and recall
contribute equally to the score. The F1 score is particularly insightful when you need a single measure to
balance precision and recall. F1 score can be calculated by (4).
∑ 𝑅𝑒𝑐𝑎𝑙𝑙∗ ∑ 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛
𝐹1 𝑆𝑐𝑜𝑟𝑒 = 2 ∗ × 100 (4)
∑ 𝑅𝑒𝑐𝑎𝑙𝑙+∑ 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛

Figure 2. Confusion matrix

3. RESULTS AND DISCUSSION

Because it is more expensive to acquire a new customer; It is five times more expensive than
retaining an existing customer [9], henceforth, there is a need to ensure the customer satisfaction which is an
important metric in every business, and a key of increasing customer lifetime value, retaining customers and
thus reducing churn. In addition, a good predictive model for customer satisfaction can help any organization
making the right decisions, improving their performance, and enhancing their market position. The subject
has been active for a long time and is still active today with multiple publications.
After reviewing all the previous algorithms, a key information comparative study was performed to
answer the following questions:
− Which AI algorithms were tested?
− What were the best algorithms and what were their score and performance?
− What metrics were employed to evaluate the suggested algorithms?
− In which sector and application domain these algorithms have been tested?
The responses to all these questions are available in the comparative study. It provides a benchmark for the
factors involved in the process of building a smart AI algorithm that works well to predict customer satisfaction.
Overall, 18 papers related to customer satisfaction and AI techniques are viewed. The benchmark
results are summarized in Table 1. While earlier studies have explored relationships between factors
influencing customer satisfaction prediction within specific domains and evaluated different AI algorithm to
determine the most effective one. This study presents a diagnosis of these researches through a benchmark to
identify AI algorithms that perform best for predicting customer satisfaction across a wide range of domains.
Table 1 summarizes the significant articles related to customer satisfaction prediction using AI
approaches. It also provides informations about the years of publication, the sector and application domain,
the algorithms utilized, the performance metrics used, the best algorithms and the best scores. Most papers
were published between 2018 and 2023. 50% of papers deal with customer satisfaction using ML algorithms,
while the other half use DL algorithms. Figure 3 represents 39% of the papers related to the banking sector,
22% papers are related to the e-commerce sector, 17% papers are linked to the Airlines industry, 17% papers

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1654-1662

Int J Artif Intell ISSN: 2252-8938  1659

are in the telecommunications sector and 5% in the retail domain. Figure 4 shows that the most used metrics
include accuracy first, then F1-score, recall, and precision.
Figure 5 indicates that among the most used ML algorithms, RF obtains the first position, then
SVM, GB, and DT. RF and GB give the best performance. Figure 6 shows that the most used DL algorithms
are CNN, LSTM, and the combination of this 2 algorithms CNN-LSTM. The optimal algorithm in terms of
performance is CNN-LSTM. We can also notice that DL models exceed ML models in most cases if these 2
types of AI algorithms are used in the same paper.

Table 1. Summary of papers related to customer satisfaction using AI

Ref. Year Algorithm Sector Application Algorithms Metrics Best Best score (%)
type domain Algorithm
[9] 2023 ML Telecom Churn prediction LR, RF, DT, Accuracy XGBOOST Accuracy: 96,2
XGBOOST F1-score F1-score: 83,6
Recall Recall: 83
Precision Precision: 54
[1] 2022 Banking Churn prediction RF, NB, GB, ET, Accuracy RF Accuracy: 86
AB, KNN, SVC, F1-score GB F1-score: 86
DT, LR, ANN,
and NB)
[2] 2022 Churn prediction LR, SVC, GBDT, Accuracy AdaBoost Accuracy: 81,1
RF, AdaBoost Recall Recall: 71,76
F1-score F1-score: 59,9
Precision Precision: 51,4
[4] 2021 Customer LR, RF, NB, Accuracy RF Accuracy: 96
satisfaction SVM, DT, KNN F1-score DT F1-score: 97
Precision LR Precision: 98
Recall Recall: 96
[25] 2020 Churn prediction KNN, DT, SVM, Accuracy RF Accuracy: 92
RF
[3] 2020 Churn prediction RF, AdaBoost, Accuracy RF Accuracy: 88,7
SVM, KNN, F1-score F1-score: 91
XGBOOST, NB, Recall Recall: 89
DT, ANN
[6] 2022 E-commerce Online shopper's NB, KNN, SVM, Accuracy GB Accuracy: 90,5
behavior LR, RF, GB, Precision Precision: 76,1
AdaBoost F1-score F1-score: 68,9
Recall Recall: 63
[7] 2020 Online shopper's DT, RF, ANN, Accuracy RF Accuracy: 87,6
behavior SVM F1-score F1-score: 93
[15] 2018 Airlines Customer NB, K-Star, IBK AUC IBK AUC: 97,59
satisfaction
[29] 2023 DL Banking CRM CNN Accuracy CNN Accuracy: 77
F1-score F1-score: 83
AUC AUC: 84
[5] 2015 Customer MultiLayer Accuracy MLP ANN Accuracy: 73
satisfaction Perceptron ANN
[16] 2022 Airlines Customer ML: KNN, DT, Accuracy CNN- Accuracy: 94
satisfaction RF, XGBoost F1-score LSTM F1-score: 93
DL: CNN, CNN- Precision Precision: 94
LSTM Recall Recall: 94
[19] 2021 Consumer ML: LR, NB, DT, Accuracy CNN- Accuracy: 91,3
Sentiment SVM, F1-score LSTM F1-score: 87,5
Analysis DL: CNN, LSTM, Precision Precision: 87,8
CNN-LSTM Recall Recall: 87
[8] 2018 E-commerce Online shopper's Multi-layered Precision ANN Precision: 80
behavior ANN
[23] 2021 Consumer ML: SVM, RF, Accuracy CNN- Accuracy: 87,5
Repurchase XGBoost Recall LSTM Recall: 83,9
behavior DL: CNN, LSTM, F1-score F1-score: 85,6
CNN-LSTM
[32] 2024 Telecom Customer LSTM Accuracy LSTM Accuracy: 95
abandonment
[31] 2023 User Satisfaction LSTM, GRU, Accuracy LSTM Accuracy: 97
Based on BiLSTM, CNN- Sensitivity Sensitivity: 93
Sentiment LSTM Specifity Specifity: 98
Analysis F1-score F1-score: 95
[27] 2022 Retail Churn prediction ANN, CNN Accuracy CNN Accuracy: 97
of Supermarket Precision Precision: 97
customer Recall Recall: 97

Artificial intelligence algorithms to predict customer satisfaction: … (Othman Berrada Chakour)

1660  ISSN: 2252-8938

Figure 3. Paper distribution by sector Figure 4. Paper distribution by used metrics

12 5
10
4
8
Frequency

Frequency
3
6
2
4
2 1

0 0
LR DT RF SVM KNN GB NB ANN CNN LSTM CNN-LSTM
ML Algorithms DL Algorithms
Used Best frequency Used Best frequency

Figure 5. Paper distribution by used ML algorithms Figure 6. Paper distribution by used DL algorithms

In our study, we discovered that many studies on predicting customer satisfaction highlight issues
with data availability and question the reliability of findings. Therefore, customer satisfaction is considered
subjective, relative, dynamic, and influenced by individual perceptions and subject to change over time,
making its definition and measurement challenging. However, further, and in-depth studies are warranted to
explore alternative data sources that could mitigates the limitations explained before, such as collecting
qualitative data from client feedback sessions or integrating data from emerging technologies like sentiment
analysis of social media posts. Future studies may explore to gather real-world data on customer satisfaction
and implement regular satisfaction surveys focusing on emotional factors. Based on our study we concluded
that the customer satisfaction’s prediction is applicable on a multitude domain. Certainly, there are some
limitations that we already listed previously in this paper, but following implications mentioned also for
future research should focus on developing robust methodologies for data cleaning and preprocessing to
enhance the reliability of predictions.

4. CONCLUSION
In summary, this paper explains the importance of customer satisfaction for any business, which
pushes decision makers to be proactive by implementing strategies based on correct predictions and close to
reality. To predict customer satisfaction, it was necessary to study the different prediction algorithms to
define the most efficient one in terms of accuracy and precision. This paper aims to find the most accurate
prediction model for customer satisfaction. The study’s results showed that RF and GB algorithms in ML and
CNN-LSTM in DL provide the best performance. In addition, DL algorithms outperform ML in most cases.
Over this study, it turns out that most articles related to the prediction of customer satisfaction report the
unavailability of data and question their reliability. Therefore, customer satisfaction is subjective, relative,
and evolving, because it depends on the individual perception of the client and can vary over time.
Satisfaction is never purely rational; it is also a matter of emotions. This is what makes it so difficult to

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1654-1662

Int J Artif Intell ISSN: 2252-8938  1661

define and measure. As next step, we intend to collect real data related to customer satisfaction, also set up
frequent satisfaction surveys considering the factors related to the customer emotions.

REFERENCES
[1] S. C. K. Tékouabou, Ș. C. Gherghina, H. Toulni, P. N. Mata, and J. M. Martins, “Towards explainable machine learning for bank
churn prediction using data balancing and ensemble-based methods,” Mathematics, vol. 10, no. 14, Jul. 2022, doi:
10.3390/math10142379.
[2] P. Chen, N. Liu, and B. Wang, “Evaluation of customer behaviour with machine learning for churn prediction: The case of bank
customer churn in Europe,” in Proceedings of the International Conference on Financial Innovation, FinTech and Information
Technology, FFIT 2022, Shenzhen, China, EAI, 2023, doi: 10.4108/eai.28-10-2022.2328450.
[3] A. Muneer, R. F. Ali, A. Alghamdi, S. M. Taib, A. Almaghthawi, and E. A. A. Ghaleb, “Predicting customers churning in
banking industry: A machine learning approach,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 26,
no. 1, pp. 539-549, Apr. 2022, doi: 10.11591/ijeecs.v26.i1.pp539-549.
[4] S. F. Shetu, I. Jahan, M. M. Islam, R. A. Hossain, N. N. Moon, and F. N. Nur, “Predicting satisfaction of online banking system in
Bangladesh by machine learning,” in 2021 International Conference on Artificial Intelligence and Computer Science Technology
(ICAICST), IEEE, Jun. 2021, pp. 223–228, doi: 10.1109/ICAICST53116.2021.9497796.
[5] N. Zeinalizadeh, A. A. Shojaie, and M. Shariatmadari, “Modeling and analysis of bank customer satisfaction using neural networks
approach,” International Journal of Bank Marketing, vol. 33, no. 6, pp. 717–732, Sep. 2015, doi: 10.1108/IJBM-06-2014-0070.
[6] V. Parihar and S. Yadav, “Comparative analysis of different machine learning algorithms to predict online shoppers’ behaviour ,”
International Journal of Advanced Networking and Applications, vol. 13, no. 06, pp. 5169–5182, 2022, doi:
10.35444/IJANA.2022.13603.
[7] A.-N. Wong and B. P. Marikannan, “Optimising e-commerce customer satisfaction with machine learning,” Journal of Physics:
Conference Series, vol. 1712, no. 1, Dec. 2020, doi: 10.1088/1742-6596/1712/1/012044.
[8] S. Agrawal, A. Das, A. Gaikwad, and S. Dhage, “Customer churn prediction modelling based on behavioural patterns analysis
using deep learning,” in 2018 International Conference on Smart Computing and Electronic Enterprise (ICSCEE), IEEE, Jul.
2018, pp. 1–6, doi: 10.1109/ICSCEE.2018.8538420.
[9] J. Maan and H. Maan, “Customer churn prediction model using explainable machine learning,” International Journal of
Computer Science Trends and Technology (IJCST), vol. 11, no. 1, pp. 33-38, Mar. 2023.
[10] S. S. Roy, D. Kaul, R. Roy, C. Barna, S. Mehta, and A. Misra, “Prediction of customer satisfaction using naive bayes, multiclass classifier,
K-star and IBK,” in Advances in Intelligent Systems and Computing, vol. 634, 2018, pp. 153–161, doi: 10.1007/978-3-319-62524-9_12.
[11] S.-H. Park, M.-Y. Kim, Y.-J. Kim, and Y.-H. Park, “A deep learning approach to analyze airline customer propensities: the case
of South Korea,” Applied Sciences, vol. 12, no. 4, Feb. 2022, doi: 10.3390/app12041916.
[12] B. K. Mohanta, D. Jena, N. Mohapatra, S. Ramasubbareddy, and B. S. Rawal, “Machine learning based accident prediction in
secure IoT enable transportation system,” Journal of Intelligent & Fuzzy Systems, vol. 42, no. 2, pp. 713–725, Jan. 2022, doi:
10.3233/JIFS-189743.
[13] T. Chuluunsaikhan, G. A. Ryu, K. H. Yoo, H. Rah, and A. Nasridinov, “Incorporating deep learning and news topic modeling for
forecasting pork prices: The case of South Korea,” Agriculture, vol. 10, no. 11, pp. 1–22, Oct. 2020, doi: 10.3390/agriculture10110513.
[14] N. Thakur and C. Y. Han, “A study of fall detection in assisted living: Identifying and improving the optimal machine learning
method,” Journal of Sensor and Actuator Networks, vol. 10, no. 3, Jun. 2021, doi: 10.3390/jsan10030039.
[15] R. Xie and D. Wu, “Optimal transport-based transfer learning for smart manufacturing: Tool wear prediction using out-of-domain
data,” Manufacturing Letters, vol. 29, pp. 104–107, Aug. 2021, doi: 10.1016/j.mfglet.2021.08.004.
[16] S. H. Park and Y. H. Park, “Audio-visual tensor fusion network for piano player posture classification,” Applied Sciences, vol. 10,
no. 19, pp. 1–15, Sep. 2020, doi: 10.3390/app10196857.
[17] C. Bhatt, I. Kumar, V. Vijayakumar, K. U. Singh, and A. Kumar, “The state of the art of deep learning models in medical science
and their challenges,” Multimedia Systems, vol. 27, no. 4, pp. 599–613, 2021, doi: 10.1007/s00530-020-00694-1.
[18] C. Lawson and D. C. Montgomery, “Logistic regression analysis of customer satisfaction data,” Quality and Reliability
Engineering International, vol. 22, no. 8, pp. 971–984, Dec. 2006, doi: 10.1002/qre.775.
[19] P. K. Jain, V. Saravanan, and R. Pamula, “A hybrid CNN-LSTM: A deep learning approach for consumer sentiment analysis
using qualitative user-generated contents,” ACM Transactions on Asian and Low-Resource Language Information Processing,
vol. 20, no. 5, pp. 1–15, Sep. 2021, doi: 10.1145/3457206.
[20] H. Gou, L. Su, G. Zhang, H. Li, and Z. Li, “Research on telecom customer satisfaction prediction strategy,” in 2023 8th
International Conference on Cloud Computing and Big Data Analytics, ICCCBDA 2023, IEEE, Apr. 2023, pp. 593–598, doi:
10.1109/ICCCBDA56900.2023.10154738.
[21] S. Ray, “A quick review of machine learning algorithms,” in 2019 International Conference on Machine Learning, Big Data,
Cloud and Parallel Computing (COMITCon), IEEE, Feb. 2019, pp. 35–39, doi: 10.1109/COMITCon.2019.8862451.
[22] A. -All-Tanvir, I. A. Khandokar, A. K. M. M. Islam, S. Islam, and S. Shatabda, “A gradient boosting classifier for purchase
intention prediction of online shoppers,” Heliyon, vol. 9, no. 4, Apr. 2023, doi: 10.1016/j.heliyon.2023.e15163.
[23] C. Zhu, M. Wang, and C. Su, “Prediction of consumer repurchase behavior based on LSTM neural network model,” International
Journal of System Assurance Engineering and Management, vol. 13, no. S3, pp. 1042–1053, 2022, doi: 10.1007/s13198-021-
01270-0.
[24] A. Mawaddah, M. H. Dar, and G. J. Yanris, “Analysis of the SVM method to determine the level of online shopping satisfaction
in the community,” SinkrOn, vol. 8, no. 2, pp. 838–855, Apr. 2023, doi: 10.33395/sinkron.v8i2.12261.
[25] M. Rahman and V. Kumar, “Machine learning based customer churn prediction in banking,” in Proceedings of the 4th
International Conference on Electronics, Communication and Aerospace Technology, ICECA 2020, IEEE, Nov. 2020,
pp. 1196–1201, doi: 10.1109/ICECA49313.2020.9297529.
[26] A. Nurdina and A. B. I. Puspita, “Naive Bayes and KNN for airline passenger satisfaction classification: comparative analysis,”
Journal of Information System Exploration and Research, vol. 1, no. 2, Jul. 2023, doi: 10.52465/joiser.v1i2.167.
[27] O. F. Seymen, E. Ölmez, O. Doğan, O. Er, and K. Hiziroğlu, “Customer churn prediction using ordinary artificial neural network
and convolutional neural network algorithms: a comparative performance assessment,” Gazi University Journal of Science,
vol. 36, no. 2, pp. 720–733, Jun. 2023, doi: 10.35378/gujs.992738.
[28] S. Pouyanfar et al., “A survey on deep learning: algorithms, techniques, and applications,” ACM Computing Surveys, vol. 51,
no. 5, pp. 1–36, Sep. 2018, doi: 10.1145/3234150.

Artificial intelligence algorithms to predict customer satisfaction: … (Othman Berrada Chakour)

1662  ISSN: 2252-8938

[29] E. M. Elgohary, M. Galal, A. Mosa, and G. A. Elshabrawy, “Smart evaluation for deep learning model: churn prediction as a
product case study,” Bulletin of Electrical Engineering and Informatics, vol. 12, no. 2, pp. 1219–1225, Apr. 2023, doi:
10.11591/eei.v12i2.4180.
[30] A. Boukhalfa, A. Abdellaoui, N. Hmina, and H. Chaoui, “LSTM deep learning method for network intrusion detection system,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 10, no. 3, pp. 3315-3322, Jun. 2020, doi:
10.11591/ijece.v10i3.pp3315-3322.
[31] M. A. Alshamari, “Evaluating user satisfaction using deep-learning-based sentiment analysis for social media data in Saudi
Arabia’s telecommunication sector,” Computers, vol. 12, no. 9, Aug. 2023, doi: 10.3390/computers12090170.
[32] S. Beltozar-Clemente, O. Iparraguirre-Villanueva, F. Pucuhuayla-Revatta, J. Zapata-Paulini, and M. Cabanillas-Carbonell,
“Predicting customer abandonment in recurrent neural networks using short-term memory,” Journal of Open Innovation:
Technology, Market, and Complexity, vol. 10, no. 1, Mar. 2024, doi: 10.1016/j.joitmc.2024.100237.
[33] R. Mutegeki and D. S. Han, “A CNN-LSTM approach to human activity recognition,” in 2020 International Conference on Artificial
Intelligence in Information and Communication, ICAIIC 2020, Feb. 2020, pp. 362–366, doi: 10.1109/ICAIIC48513.2020.9065078.
[34] E. Ismanto, H. Ab. Ghani, and N. I. B. Md Saleh, “A comparative study of machine learning algorithms for virtual learning
environment performance prediction,” IAES International Journal of Artificial Intelligence (IJ-AI), vol. 12, no. 4, pp. 1677-1686,
Dec. 2023, doi: 10.11591/ijai.v12.i4.pp1677-1686.

BIOGRAPHIES OF AUTHORS

Othman Berrada Chakour holds an engineering degree in software engineering

from ENSIAS, Mohamed V University, Rabat in 2014 and a bachelor’s degree in mathematics
and computer science from Hassan II University, Casablanca in 2011. His research includes
customer satisfaction prediction via machine learning and deep learning. He can be contacted
at email: [email protected] or [email protected].

Dr. Abdelaziz Ettaoufik is Professor in the Department of Mathematics and

Informatics at Faculty of Sciences Ben M’Sik, Hassan II University, Morocco. His research
interest includes computer sciences, big data & cloud computing, intelligence artificial, data
warehouse optimization, data lake, and blockchain. He can be contacted at email:
[email protected].

Dr. Khalid Aissaoui completed his Ph.D. at University Hassan II/ENSEM

Casablanca Morocco in 2020. Received an engineering degree in 2008 at ENSSAT- Lannion
France. Working as Professor Researcher at higher institute of applied engineering IGA,
Casablanca Morocco. His research focuses on security & IAM in cloud computing, big data,
and artificial intelligence. He can be contacted at email: [email protected].

Dr. Abderrahim Maizate obtained his engineering diploma in computer science

from the Hassania School of Public Works in 2004, followed by a DESA degree from
ENSIAS in 2007. He completed his Ph.D. at Chouaib Doukkali University. Currently, he
serves as a Professor Researcher at Hassan II University in Casablanca, Morocco. His research
focuses on wireless communication, mobile communication, wireless sensor networks, quality
of service (QoS) guarantees, and big data. He can be contacted at email:
[email protected] or [email protected].

Int J Artif Intell, Vol. 14, No. 2, April 2025: 1654-1662