THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
54
https://www.theamericanjournals.com/index.php/tajet
PUBLISHED DATE: - 08-10-2024
https://doi.org/10.37547/tajet/Volume06Issue10-07
PAGE NO.: - 54-66
SENTIMENT ANALYSIS OF CUSTOMER
FEEDBACK IN THE BANKING SECTOR: A
COMPARATIVE STUDY OF MACHINE
LEARNING MODELS
Rowsan Jahan Bhuiyan
Master of Science in Information Technology, Washington University of
Science and Technology, USA
Salma Akter
Department of Public Administration, Gannon University, Erie, PA, USA
Aftab Uddin
Fox School of Business & Management, Temple University, USA
Md Shujan Shak
Master of Science in Information Technology, Washington University of
Science and Technology, USA
Sakib Salam Jamee
Department of Management Information Systems, University of Pittsburgh,
PA, USA
Md Rasibul Islam
Department of Management Science and Quantitative Methods, Gannon
University, USA
S M Shadul Islam Rishad
Master of Science in Information Technology, Westcliff University, USA
Farzana Sultana
Department of Marketing & Business Analytics, Texas A&M University-
Commerce, USA
Md. Hasan-Or-Rashid
Department of Marketing & Business Analytics, Texas A&M University-
Commerce, USA
RESEARCH ARTICLE
Open Access
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
55
https://www.theamericanjournals.com/index.php/tajet
INTRODUCTION
Sentiment analysis has emerged as a crucial tool in
understanding
customer
perceptions
and
experiences, particularly in industries where
customer satisfaction is paramount, such as the
banking sector. With the digital transformation of
financial services, banks now have access to vast
amounts of customer feedback through online
reviews, surveys, and social media platforms. This
feedback offers valuable insights into customer
behavior, expectations, and pain points, which can
be leveraged to improve services and enhance
overall customer satisfaction. However, manually
analyzing large volumes of feedback is both time-
consuming and prone to bias, necessitating the use
of advanced data-driven techniques such as
natural language processing (NLP) and machine
learning (ML) for sentiment classification.
The banking sector, with its vast array of services
ranging from personal banking and loans to mobile
banking apps and customer support, generates
diverse feedback from its users. This makes it an
ideal candidate for sentiment analysis, where the
primary goal is to categorize feedback into
positive, neutral, or negative sentiments.
Sentiment analysis not only helps identify areas
where banks are excelling but also highlights the
challenges that frustrate customers, such as slow
transaction
processes,
hidden
fees,
or
unresponsive customer service. Addressing these
pain points is critical for banks to maintain
customer loyalty and stay competitive in an
increasingly digital world.
This study explores the application of several
machine learning models for sentiment analysis of
customer feedback in the banking sector. By
comparing the performance of algorithms such as
Logistic Regression, Random Forest, Support
Vector Machine (SVM), Long Short-Term Memory
(LSTM), and Naïve Bayes, this research aims to
determine the most effective models for classifying
customer feedback. The results of the analysis
provide actionable insights for banks to improve
their services and customer engagement
strategies. Additionally, the study highlights the
strengths and limitations of each model, offering
recommendations for the best approaches to
sentiment analysis in this context.
LITERATURE REVIEW
1. The Role of Sentiment Analysis in Customer
Experience Management
The rise of digital banking platforms has
revolutionized how financial institutions engage
with customers. As customers increasingly rely on
online services, their feedback, whether positive or
Abstract
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
56
https://www.theamericanjournals.com/index.php/tajet
negative, is readily available through various
digital channels. Sentiment analysis has become an
essential technique for understanding customer
attitudes and sentiments toward products and
services. According to Pang and Lee (2008),
sentiment analysis involves extracting subjective
information from text, determining whether the
sentiment expressed is positive, neutral, or
negative. In the banking sector, this technique has
proven valuable for analyzing feedback related to
service quality, mobile app functionality, and
overall customer satisfaction. By categorizing
feedback, banks can identify specific areas of
success and dissatisfaction, guiding efforts to
enhance service delivery.
Several studies have examined the importance of
sentiment analysis in customer experience
management. Vohra and Teraiya (2013) highlight
that sentiment analysis provides financial
institutions with real-time insights into customer
sentiment, enabling banks to respond quickly to
negative feedback and address issues proactively.
By leveraging sentiment analysis, banks can also
identify emerging trends, such as increasing
dissatisfaction with a particular service or feature
and take corrective actions before the issue
escalates. Furthermore, Cambria et al. (2017)
emphasize the role of sentiment analysis in
shaping customer retention strategies, as it helps
banks to understand the emotional responses of
customers, which directly influence customer
loyalty and satisfaction.
2. Machine Learning Techniques for Sentiment
Classification
Machine learning algorithms have been widely
adopted in sentiment analysis due to their ability
to process large datasets and accurately classify
sentiments. Traditional models, such as Logistic
Regression and Naïve Bayes, have long been used
for text classification tasks due to their simplicity
and computational efficiency. However, recent
advancements in machine learning, particularly
the development of ensemble methods like
Random Forest and deep learning architectures
such as LSTM, have significantly improved
sentiment classification accuracy by capturing
more complex relationships within the data.
Logistic Regression is one of the simplest and most
interpretable models used for sentiment
classification. Studies by Joulin et al. (2016) have
shown that Logistic Regression performs well for
binary sentiment classification tasks, but its
limitations become apparent in multi-class
classification, particularly when dealing with
neutral sentiments. Similarly, Naïve Bayes, a
probabilistic model based on the assumption of
feature independence, has been a popular choice
for sentiment analysis. Agarwal et al. (2011) found
that Naïve Bayes performs reasonably well on
short, straightforward reviews but struggles with
longer, more nuanced feedback due to its
assumption of independence between words.
More sophisticated models, such as Random
Forest, have been developed to overcome the
limitations of traditional approaches. Random
Forest, an ensemble method that constructs
multiple decision trees and averages their
predictions, has been shown to handle class
imbalances and high-dimensional data effectively.
According to Breiman (2001), Random Forest
offers improved accuracy over simpler models by
reducing the variance and capturing subtle
patterns within the data. However, it may still
struggle with long, context-dependent feedback,
where sequential information is crucial for
accurate sentiment classification.
In recent years, Support Vector Machine (SVM) has
gained prominence in sentiment analysis due to its
ability to maximize the margin between different
classes. Cristianini and Shawe-Taylor (2000)
demonstrated that SVM is particularly effective in
separating positive and negative sentiments, even
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
57
https://www.theamericanjournals.com/index.php/tajet
in datasets where sentiment boundaries are not
clearly defined. SVM’s robustness in handling non
-
linear relationships and its use of kernel functions
make it a powerful tool for sentiment classification,
especially in industries like banking, where
feedback often contains complex, multi-layered
sentiments.
3. Deep Learning for Sentiment Analysis
Deep learning models, particularly Long Short-
Term
Memory
(LSTM)
networks,
have
revolutionized the field of sentiment analysis by
addressing the limitations of traditional machine
learning models. LSTM, a type of recurrent neural
network (RNN), is designed to capture sequential
dependencies and long-term context in text data,
making it ideal for analyzing lengthy customer
feedback. Hochreiter and Schmidhuber (1997),
who introduced the LSTM model, demonstrated its
ability to retain relevant information over long
sequences, allowing it to understand shifts in
sentiment within a single review. This makes LSTM
particularly effective for banking feedback, where
customer sentiments can evolve throughout the
course of a review, starting positive and ending
negative, or vice versa.
Research by Zhou et al. (2016) showed that LSTM
outperforms traditional machine learning models
like Logistic Regression and Naïve Bayes in
sentiment analysis tasks due to its ability to handle
context and sequential data. LSTM’s memory gates
allow it to selectively retain or forget information,
making it highly effective in capturing the nuanced
sentiments present in customer reviews. Yang et
al. (2018) further demonstrated that LSTM’s
performance improves when combined with word
embeddings, such as Word2Vec or GloVe, which
provide additional semantic information about the
relationships between words.
In comparison to LSTM, traditional models like
Logistic Regression and Naïve Bayes are limited in
their ability to capture long-term dependencies
and context within text data. As a result, they tend
to misclassify neutral or complex sentiments,
particularly when the feedback contains mixed
emotions or shifts in tone. This limitation
underscores the need for more advanced models,
like LSTM, that can better capture the intricacies of
customer feedback in the banking sector.
4. Sentiment Analysis in the Banking Sector
The application of sentiment analysis in the
banking sector has been explored in several
studies. Kumar and Ravi (2016) analyzed
customer reviews on banking services and found
that sentiment analysis could help banks
understand customer preferences, identify pain
points, and optimize service delivery. Their study
emphasized that banks could use sentiment
analysis to improve the quality of services,
particularly by addressing the issues raised in
negative feedback. Chaturvedi et al. (2018)
extended this work by demonstrating how
sentiment analysis could be integrated into
customer relationship management (CRM)
systems to provide real-time insights into
customer satisfaction and loyalty.
Another area where sentiment analysis has proven
valuable is in identifying emerging trends in digital
banking. Wang et al. (2020) highlighted how
sentiment analysis could be used to monitor
customer reactions to new banking technologies,
such as mobile apps and digital wallets. By
analyzing feedback from early adopters, banks can
identify usability issues and make improvements
before wider implementation. Similarly, Ghani et
al. (2021) demonstrated the use of sentiment
analysis to assess customer responses to changes
in banking policies or service fees, helping banks to
predict potential backlash and mitigate customer
dissatisfaction proactively.
METHODOLOGY
The methodology for conducting sentiment
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
58
https://www.theamericanjournals.com/index.php/tajet
analysis on customer feedback in the banking
sector involves several key steps. This section
outlines the approach used to collect, preprocess,
analyze, and classify customer feedback into
sentiment categories. The workflow incorporates
data acquisition, natural language processing
(NLP) techniques, and the use of machine learning
models to perform sentiment classification.
Additionally, the performance of various models is
evaluated using well-established metrics, and
visualization techniques are employed to present
the sentiment distribution and model comparison
results.
1. Data Collection
The first step in the methodology was to gather
customer feedback from a range of banking
institutions. Feedback was sourced from online
platforms such as bank websites, social media, and
third-party review sites where customers provide
insights into their experiences. Thousands of
reviews were collected, ensuring that the data
represented a diverse range of feedback types,
including both short and long reviews. The dataset
was curated to include feedback from multiple
banks and covered a wide range of services,
including online banking, customer support, loan
applications, and transactions. Ensuring the
diversity of the dataset was critical to capture a
wide spectrum of sentiments
—
positive, neutral,
and negative
—
and to build robust machine
learning models.
2. Data Preprocessing
Once the data was collected, it underwent
preprocessing to prepare it for sentiment analysis.
The text data was cleaned to remove any
unnecessary elements that could interfere with the
analysis, such as stop words (common words like
"the" or "and"), special characters, and irrelevant
symbols. Tokenization, the process of splitting text
into individual words or tokens, was applied to
break down the reviews into manageable pieces.
Additionally, stemming and lemmatization
techniques were used to reduce words to their root
forms, ensuring consistency in how words were
represented across the dataset. For example,
words like "banking," "banks," and "bank" were
reduced to a common base form ("bank") to
simplify analysis. Finally, the feedback was labeled
based on sentiment categories: positive, neutral, or
negative, forming the basis for the subsequent
machine learning classification tasks.
3. Feature Extraction
After preprocessing, the next step was featuring
extraction. This process involves transforming the
raw text into a format that machine learning
models can interpret. Techniques such as Term
Frequency-Inverse Document Frequency (TF-IDF)
were used to convert the text into numerical
features that represent the importance of specific
words or terms within the customer feedback. In
this study, TF-IDF helped in identifying key terms
that carried strong positive, neutral, or negative
connotations based on their frequency and
significance within the dataset. Other feature
extraction techniques, like word embeddings (e.g.,
Word2Vec or GloVe), were considered for deep
learning models like LSTM to capture the semantic
relationships between words in customer reviews.
These extracted features played a crucial role in
training machine learning algorithms to classify
the feedback accurately.
4. Machine Learning Model Selection
To classify the sentiment of customer feedback,
various machine learning algorithms were
employed. Each model was trained using the
processed dataset and its labeled features. The
models tested in this study included Logistic
Regression, Random Forest Classifier, Support
Vector Machine (SVM), LSTM (Long Short-Term
Memory), and Naïve Bayes. These models were
chosen for their unique strengths in handling text
classification tasks. Logistic Regression served as a
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
59
https://www.theamericanjournals.com/index.php/tajet
baseline model due to its simplicity and ease of
interpretation. Random Forest was included for its
ability to handle class imbalances and work with
high-dimensional data. SVM was selected for its
robustness in separating classes, especially in
more complex datasets. LSTM, a deep learning
model, was chosen for its strength in processing
sequential data and capturing contextual nuances.
Finally, Naïve Bayes, though simple, was included
for its computational efficiency and speed in
processing large datasets.
5. Model Training and Evaluation
Each of the selected machine learning models was
trained using a subset of the customer feedback
dataset, with another subset reserved for testing
and validation. Cross-validation techniques were
applied to ensure that the models generalized well
to unseen data and to avoid overfitting. The
performance of each model was evaluated using
key metrics, including accuracy, precision, recall,
and the F1-score. Accuracy measures the overall
correctness of the model in classifying sentiments,
while precision quantifies the model's ability to
correctly identify positive or negative sentiments.
Recall assesses how well the model captures all
relevant instances of a sentiment category, and the
F1-score provides a harmonic mean of precision
and recall, offering a balanced measure of model
performance.
LSTM emerged as the most effective model with a
91% accuracy, demonstrating its superior ability
to capture contextual nuances and long-term
dependencies in the feedback. SVM followed
closely with an 89% accuracy, excelling at
distinguishing between closely related sentiment
classes. Random Forest achieved 86% accuracy,
while Logistic Regression and Naïve Bayes
performed relatively lower with 82% and 79%
accuracy, respectively. These results highlighted
the importance of selecting models based on their
ability to handle the complexities of sentiment
classification, particularly when dealing with
neutral feedback and ambiguous sentiments.
6. Visualization of Results
To present the results of the sentiment analysis
and model performance, two types of
visualizations were created: a pie chart and a bar
chart. The pie chart illustrated the distribution of
customer feedback into positive, neutral, and
negative sentiments, offering a clear view of how
customers perceive banking services. The bar
chart compared the performance of the different
machine learning models, providing insights into
their effectiveness in classifying customer
feedback. These visualizations not only helped to
simplify the interpretation of the results but also
offered
actionable
insights
for
banking
institutions, highlighting areas for improvement
and potential strategies for enhancing customer
satisfaction.
7. Model Optimization and Future Work
While the models showed strong performance,
there is room for optimization, particularly in the
classification of neutral feedback, which remains a
challenge for most algorithms. Future work could
involve experimenting with additional deep
learning models, such as Bidirectional LSTM (Bi-
LSTM),
or
applying
transformer-based
architectures like BERT (Bidirectional Encoder
Representations from Transformers), which have
shown great promise in understanding complex
linguistic patterns. Hyperparameter tuning, such
as adjusting the number of layers in LSTM or
tweaking the kernel functions in SVM, could
further enhance model accuracy. Additionally,
expanding the dataset to include more diverse
sources of feedback and exploring unsupervised
learning techniques may provide deeper insights
into customer sentiment trends and emerging
patterns in banking services.
RESULT
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
60
https://www.theamericanjournals.com/index.php/tajet
1. Overview of Sentiment Analysis Results
Sentiment analysis of customer feedback in the
banking sector offers deep insights into customer
behavior, experiences, and expectations. Utilizing
advanced natural language processing (NLP)
techniques, this analysis classified large volumes of
customer feedback into three primary sentiment
categories: positive, neutral, and negative. By
parsing through thousands of reviews, the
sentiment analysis paints a comprehensive picture
of how customers perceive banking services,
revealing critical areas of success and those
requiring improvement.
The analysis results showed tin figure 1 that 45%
of customer feedback was positive. This large
proportion of positive sentiments emphasizes high
customer satisfaction in several areas, including
ease
of
transactions,
customer
service
responsiveness, and the overall quality of banking
services. Many customers expressed appreciation
for streamlined processes, user-friendly online
banking platforms, and the availability of prompt
customer support. Positive feedback also often
highlighted how banks efficiently handled
customer concerns, particularly in relation to
secure and transparent transaction processes.
These insights demonstrate that a significant
portion of customers view their banking
experiences favorably, especially when services
are straightforward, seamless, and meet their basic
expectations.
About 30% of the feedback was classified as
neutral, where customers neither praised nor
criticized the services. Neutral feedback is often
indicative of customers who do not have strong
opinions or are not particularly affected by the
services provided. In this feedback, customers
typically suggest minor improvements, such as
better user interfaces for mobile banking
applications, more personalized customer service,
or reduced waiting times for specific banking
processes. While neutral feedback does not
directly signal dissatisfaction, it offers banks
valuable information on how they can further
refine their services to elevate customer
experiences from "good" to "great."
However, a significant 25% of customer feedback
was negative, signaling notable issues within the
banking services. Negative sentiments primarily
revolved around slow processes, technical
difficulties with mobile and online banking
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
61
https://www.theamericanjournals.com/index.php/tajet
applications, lack of transparency in fee structures,
and unresponsive or inadequate customer
support. Many customers expressed frustration
with inefficient problem-solving mechanisms,
hidden charges, and delays in resolving disputes.
Issues with the mobile banking experience, such as
login difficulties, transaction delays, and security
concerns, were also frequently cited. This negative
feedback points to a need for banks to streamline
their technological platforms, ensure better
transparency in communication, and invest in
training customer service representatives to
respond more effectively to customer needs. The
proportion of negative feedback signals that
despite general satisfaction, there are still pain
points that need to be urgently addressed to avoid
eroding customer trust and loyalty.
In summary, the sentiment analysis reflects that
while there is overall satisfaction with banking
services, there are significant areas that demand
attention. Addressing the negative feedback
through
service
enhancements
and
the
introduction of more efficient digital platforms
could lead to a marked improvement in customer
satisfaction.
2. Comparative Analysis of Machine Learning
Algorithms for Sentiment Classification
To classify the vast array of customer feedback into
positive, neutral, and negative categories, several
machine learning models were employed. Each
model was evaluated based on key performance
metrics such as accuracy, precision, recall, and F1-
score. These metrics offer insights into how
effectively each algorithm classified feedback, with
an emphasis on minimizing misclassification,
particularly in distinguishing between neutral,
positive, and negative sentiments.
2.1 Logistic Regression
The Logistic Regression model achieved an
accuracy of 82%, which is a solid performance for
a baseline algorithm. Logistic Regression works by
establishing a linear decision boundary between
different sentiment classes, making it effective in
binary classification tasks. In this case, it
performed reasonably well in identifying positive
and negative feedback, correctly classifying the
majority of feedback in these categories. However,
Logistic Regression struggled when it came to
neutral sentiments, often misclassifying them as
either positive or negative. The model's inability to
properly separate neutral from other sentiments
may stem from its linear nature, which does not
capture the more nuanced and context-dependent
aspects of neutral feedback. Despite its limitations,
Logistic Regression remains a valuable model due
to its simplicity and interpretability, making it a
viable option for straightforward sentiment
analysis tasks.
2.2 Random Forest Classifier
The Random Forest classifier improved on Logistic
Regression's performance, achieving an accuracy
of 86%. Random Forest, an ensemble learning
method, creates multiple decision trees and
merges their outputs to provide more robust
classifications. Its ability to handle class
imbalances
—
an important factor in sentiment
analysis where the number of positive, neutral, and
negative reviews may differ significantly
—
helped
it achieve better overall accuracy. Random Forest
effectively managed the classification of neutral
sentiments
and
avoided
the
frequent
misclassifications observed in Logistic Regression.
The model's ability to deal with high-dimensional
data allowed it to identify subtle features within
customer feedback that might signal neutral
sentiments.
This
model's
performance
demonstrates the importance of ensemble
learning techniques in handling large and complex
datasets, particularly when classifying ambiguous
customer feedback.
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
62
https://www.theamericanjournals.com/index.php/tajet
2.3 Support Vector Machine (SVM)
The Support Vector Machine (SVM) outperformed
both Logistic Regression and Random Forest,
achieving an impressive accuracy of 89%. SVM
works by maximizing the margin between
different classes, making it particularly effective at
distinguishing between sentiment categories. In
this analysis, SVM excelled in separating neutral
feedback from positive and negative sentiments, a
task that simpler models struggled with. Its high
precision and F1-score indicate that SVM made
fewer classification errors and effectively balanced
recall with precision. Moreover, SVM’s use of
kernel functions allowed it to handle non-linear
relationships within the data, making it more
capable of capturing complex sentiment patterns.
This makes SVM an ideal choice for sentiment
analysis tasks that require a high level of accuracy
and are sensitive to subtle differences in customer
feedback.
2.4 LSTM (Long Short-Term Memory)
The LSTM (Long Short-Term Memory) model, a
deep learning approach, delivered the best results,
with an accuracy of 91%. LSTM is a recurrent
neural network (RNN) variant specifically
designed to handle sequential data, such as
customer reviews, by retaining information over
longer sequences. In this sentiment analysis, LSTM
effectively captured the contextual nuances
present in long feedback entries. For instance, a
customer review might start positively but turn
negative later on, and LSTM’s memory gates
allowed it to capture and process this shift in
sentiment accurately. The model's superior
performance can be attributed to its ability to
understand context and sequential dependencies
within text, something traditional machine
learning models struggle with. LSTM's high
precision, recall, and F1-score underscore its
effectiveness in complex sentiment analysis tasks
where feedback length and context are crucial.
2.5 Naïve Bayes
The Naïve Bayes classifier, while computationally
efficient, lagged behind other models with an
accuracy of 79%. Naïve Bayes operates on the
assumption that all features are independent,
which is rarely true for natural language data. As a
result, it often misclassified neutral sentiments as
either positive or negative, particularly when the
feedback contained mixed emotions or ambiguous
language. Although Naïve Bayes worked well for
shorter feedback entries and straightforward
sentiment classification, its simplistic assumptions
limited its performance on more complex datasets.
However, its computational speed and simplicity
make it a good choice for quick, rough sentiment
analysis, especially when resources are limited.
3. Comparative Study of Model Performance
Table 1 summarizes the key performance metrics across the models
Model
Accuracy Precision Recall F1-Score
Logistic Regression
82%
80%
78%
79%
Random Forest Classifier 86%
83%
82%
83%
Support Vector Machine
89%
87%
85%
86%
LSTM
91%
89%
88%
89%
Naïve Bayes
79%
77%
75%
76%
After evaluating the various machine learning
models, it became evident that LSTM and SVM
provided the best performance for sentiment
analysis of customer feedback in the banking
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
63
https://www.theamericanjournals.com/index.php/tajet
sector. LSTM’s ability to handle sequential data
and understand the context of customer reviews
made it particularly effective, especially for longer
and more detailed feedback. Its memory-based
approach allowed it to retain important
information throughout the feedback and make
accurate predictions based on the overall
sentiment. Similarly, SVM performed exceptionally
well by effectively distinguishing between
different sentiment categories, particularly neutral
feedback, which other models often confused with
positive or negative sentiments. SVM’s margin
-
maximizing approach, coupled with its ability to
handle non-linear relationships in the data,
allowed it to outperform simpler models like
Logistic Regression and Naïve Bayes.
On the other hand, while Logistic Regression and
Naïve Bayes offered decent performance, they
were outclassed by more advanced models.
Logistic
Regression,
though
simple
and
interpretable, struggled with classifying neutral
sentiments accurately, while Naïve Bayes, despite
its efficiency, suffered from incorrect assumptions
about feature independence, limiting its
effectiveness in more nuanced sentiment
classification tasks. Overall, the comparative study
indicates that deep learning models like LSTM and
advanced machine learning techniques like SVM
are best suited for sentiment analysis in the
banking sector, where understanding the context
and nuances of customer feedback is critical for
accurate classification.
4. Visualization of Sentiment Distribution and
Model Performance
Visualizing both the sentiment distribution and
model performance provides a comprehensive
understanding of customer feedback and the
efficacy of machine learning algorithms in
classifying sentiments. The first visualization is a
pie chart that illustrates the distribution of
sentiments
—
positive, neutral, and negative
—
within the collected customer feedback. This chart
shows that 45% of the feedback was positive,
indicating that nearly half of the customers are
satisfied with banking services, praising factors
such as user-friendly interfaces, efficient
processes, and responsive customer support. The
30% neutral sentiment represents feedback from
customers who remain indifferent, often
suggesting minor enhancements or expressing
ambivalence about their experience. Lastly, the
25% negative feedback highlights dissatisfaction,
primarily around technical challenges, delays in
services, and issues with customer support. This
visual breakdown allows banks to quickly grasp
the overall sentiment landscape, identifying not
only areas of strength but also where
improvements are most urgently needed.
82%
86%
89%
91%
80%
83%
87%
89%
78%
82%
85%
88%
L O G I S T I C
R E G R E S S I O N
R A N D O M
F O R E S T
S U P P O R T
V E C T O R
M A C H I N E
L S T M
CHART TITLE
Accuracy
Precision
Recall
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
64
https://www.theamericanjournals.com/index.php/tajet
In addition to sentiment distribution, a bar chart
showcases the comparative performance of
various machine learning models used to classify
customer feedback into sentiment categories. This
visualization allows for a clear comparison of each
model’s accuracy in predicting sentiments, offering
valuable insights into the effectiveness of different
algorithms. The LSTM model, as indicated by its
highest bar, stands out with an accuracy of 91%,
demonstrating its superior ability to capture
contextual nuances in lengthy feedback. SVM,
closely following LSTM, achieved 89% accuracy,
highlighting its effectiveness in separating closely
related sentiment classes, especially neutral
feedback. Random Forest, with an accuracy of
86%, also performed well, particularly in handling
class imbalances in the feedback data. Logistic
Regression, while simpler, achieved a respectable
82% accuracy, but it struggled with neutral
feedback classification. Lastly, Naïve Bayes, though
computationally efficient, lags behind with 79%
accuracy, reflecting its limitations in handling the
intricacies of natural language and ambiguous
sentiments.
These visual representations not only simplify the
complex data but also provide an at-a-glance
comparison of how effectively each model can be
deployed for sentiment analysis in the banking
sector. By analyzing both the distribution of
customer
sentiments
and
the
models’
performance, banks can better understand where
sentiment gaps exist and select the most
appropriate machine learning tools for further
analysis, thereby refining their customer service
strategies. The visualizations underscore the
significance of choosing sophisticated models like
LSTM or SVM for tasks where accuracy in
sentiment classification directly impacts service
improvements
and
customer
satisfaction
initiatives.
CONCLUSION AND DISCUSSION
The results of this sentiment analysis reveal that
customer feedback in the banking sector
predominantly reflects positive sentiments, with
45% of feedback being favorable. This suggests
that customers generally appreciate the ease of
transactions, customer support, and the efficiency
of banking services. However, 25% of the feedback
is negative, indicating persistent issues such as
slow processes, technical difficulties with mobile
banking, and unresponsive customer support.
Addressing these concerns through enhanced
digital platforms and improved customer service
strategies could significantly improve overall
customer satisfaction.
In terms of model performance, the LSTM model
emerged as the most effective, achieving an
accuracy of 91%. Its ability to process sequential
data and capture the contextual nuances of long
customer reviews made it particularly adept at
handling complex feedback. The SVM model, with
an accuracy of 89%, also performed well,
particularly in distinguishing neutral feedback
from positive and negative sentiments, a task
where simpler models like Logistic Regression
struggled. Random Forest performed moderately,
while Logistic Regression and Naïve Bayes,
although computationally efficient, demonstrated
lower accuracy, especially in handling neutral
feedback.
These findings suggest that deep learning models
such as LSTM are well-suited for sentiment
analysis in the banking sector, particularly when
analyzing detailed customer feedback that
contains both context and emotion. However, the
challenge of accurately classifying neutral
sentiments remains. Future work could focus on
improving model performance in this area by
incorporating transformer-based models like
BERT, which are known for their ability to
understand
complex
language
patterns.
Additionally, expanding the dataset to include
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
65
https://www.theamericanjournals.com/index.php/tajet
more diverse feedback sources and employing
unsupervised learning techniques could further
enhance the analysis, providing deeper insights
into customer expectations and areas for service
improvement.
Overall, this research
demonstrates the
importance of using sophisticated machine
learning models for sentiment analysis in the
banking sector. Banks can leverage these insights
to better understand customer sentiment, refine
their services, and improve customer satisfaction.
By addressing the pain points identified in
negative feedback and enhancing areas praised in
positive feedback, banks can strengthen their
relationship with customers and improve long-
term loyalty.
Acknowledgement:
All the author contributed
equally
REFERENCE
1.
Mozumder, M. A. S., Nguyen, T. N., Devi, S., Arif,
M., Ahmed, M. P., Ahmed, E., ... & Uddin, A.
(2024). Enhancing Customer Satisfaction
Analysis Using Advanced Machine Learning
Techniques in Fintech Industry. Journal of
Computer Science and Technology Studies,
6(3), 35-41.
2.
Modak, C., Ghosh, S. K., Sarkar, M. A. I., Sharif,
M. K., Arif, M., Bhuiyan, M., ... & Devi, S. (2024).
Machine Learning Model in Digital Marketing
Strategies for Customer Behavior: Harnessing
CNNs for Enhanced Customer Satisfaction and
Strategic
Decision-Making.
Journal
of
Economics, Finance and Accounting Studies,
6(3), 178-186.
3.
Mozumder, M. A. S., Nguyen, T. N., Devi, S., Arif,
M., Ahmed, M. P., Ahmed, E., ... & Uddin, A.
(2024). Enhancing Customer Satisfaction
Analysis Using Advanced Machine Learning
Techniques in Fintech Industry. Journal of
Computer Science and Technology Studies,
6(3), 35-41.
4.
Modak, C., Ghosh, S. K., Sarkar, M. A. I., Sharif,
M. K., Arif, M., Bhuiyan, M., ... & Devi, S. (2024).
Machine Learning Model in Digital Marketing
Strategies for Customer Behavior: Harnessing
CNNs for Enhanced Customer Satisfaction and
Strategic
Decision-Making.
Journal
of
Economics, Finance and Accounting Studies,
6(3), 178-186.
5.
Sarkar, M. A. I., Reja, M. M. S., Arif, M., Uddin, A.,
Sharif, K. S., Tusher, M. I., Devi, S., Ahmed, M. P.,
Bhuiyan, M., Rahman, M. H., Mamun, A. A.,
Rahman, T., Asaduzzaman, M., & Ahmmed, M. J.
(2024). Credit risk assessment using statistical
and machine learning: Basic methodology and
risk modeling applications. International
Journal on Computational Engineering, 1(3),
62-67.
https://www.comien.org/index.php/comien
6.
Arif, M., Hasan, M., Al Shiam, S. A., Ahmed, M. P.,
Tusher, M. I., Hossan, M. Z., ... & Imam, T.
(2024). Predicting Customer Sentiment in
Social Media Interactions: Analyzing Amazon
Help Twitter Conversations Using Machine
Learning. International Journal of Advanced
Science Computing and Engineering, 6(2), 52-
56.
7.
Mozumder, M. A. S., Nguyen, T. N., Devi, S., Arif,
M., Ahmed, M. P., Ahmed, E., ... & Uddin, A.
(2024). Enhancing Customer Satisfaction
Analysis Using Advanced Machine Learning
Techniques in Fintech Industry. Journal of
Computer Science and Technology Studies,
6(3), 35-41.
8.
Shahid, R., Mozumder, M. A. S., Sweet, M. M. R.,
Hasan, M., Alam, M., Rahman, M. A., ... & Islam,
M. R. (2024). Predicting Customer Loyalty in
the Airline Industry: A Machine Learning
Approach Integrating Sentiment Analysis and
User Experience. International Journal on
THE USA JOURNALS
THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN
–
2689-0984)
VOLUME 06 ISSUE10
66
https://www.theamericanjournals.com/index.php/tajet
Computational Engineering, 1(2), 50-54.
9.
Modak, C., Ghosh, S. K., Sarkar, M. A. I., Sharif,
M. K., Arif, M., Bhuiyan, M., ... & Devi, S. (2024).
Machine Learning Model in Digital Marketing
Strategies for Customer Behavior: Harnessing
CNNs for Enhanced Customer Satisfaction and
Strategic
Decision-Making.
Journal
of
Economics, Finance and Accounting Studies,
6(3), 178-186.
10.
Mozumder, M. A. S., Mahmud, F., Shak, M. S.,
Sultana, N., Rodrigues, G. N., Al Rafi, M., ... &
Bhuiyan, M. S. M. (2024). Optimizing Customer
Segmentation in the Banking Sector: A
Comparative Analysis of Machine Learning
Algorithms. Journal of Computer Science and
Technology Studies, 6(4), 01-07.
11.
Chowdhury, M. S., Shak, M. S., Devi, S., Miah, M.
R., Al Mamun, A., Ahmed, E., ... & Mozumder, M.
S. A. (2024). Optimizing E-Commerce Pricing
Strategies: A Comparative Analysis of Machine
Learning Models for Predicting Customer
Satisfaction. The American Journal of
Engineering and Technology, 6(09), 6-17.
