public's opinion on a particular topic. Users have unrestricted access to this website and can express their

opinions freely without any restrictions, and it is well-known that opinions influence readers. Therefore, the main

objective of this research is to identify the public's positive, negative, and neutral attitudes towards airlines such

as Malaysian Airlines, Air Asia, and Malindo Air. Two approaches are adopted: the lexicon-based approach to

label the tweets and the machine learning approach such as Naïve Bayes, SVM, and Deep Learning to predict

and compare the performance. A total of 35,005 tweets from airlines with all three keywords were evaluated.

Deep Learning achieved the highest accuracy and f1 score with 74.10% and 73.49%, respectively. The results

show that Deep Learning outperforms the other classifiers by having the highest precision and f1 score. Finally,

Microblogging websites have become a source of a

wide range of information due to the nature of

microblogs, where individuals post their thoughts on

various topics in real time, discuss current issues,

offer criticism, and express positive emotions about

things they use every day. Millions of travellers share

their opinions and viewpoints on airlines, facilities,

and services provided on social media platforms such

as Twitter, Facebook, and blogs due to the

organizations, creates new research opportunities in

much more effective in capturing the general public's

sentiments, especially in the airline industry, where

people are easily satisfied or dissatisfied and often

express their feelings on Twitter [6]. It consists of

many approaches: abstract mining, fine-grained

sentiment analysis, emotion recognition, and

multilingual sentiment analysis [5].

Air transport is one of the fastest ways to travel

worldwide. Malaysia has renowned airlines such as

Air Asia, Malaysian Airlines, and Malindo Air. With

that the airline industry is confronted with various

aspects, including passengers' needs, wants, and

comfort.

The rest of this paper is structured as follows: Section

2 provides the related works, and Section 3 presents

the literature about predictive models. Section 4

introduces the methods and dataset used in this study.

The results and findings are given in Section 5.

Lastly, Section 6 concludes the study.

usually have to gain experience using the stuff and

can only evaluate whether it is worth committing to

or vice versa. Moreover, the reviews they acquired

from friends are most likely biased, so they need to

determine whether the reviews are authentic, work

well for them, and are justifiable. However, with the

emergence of the Internet and the Web, people can

easily scrape information and data from strangers and

gather that information and reviews to evaluate

whether they are worth having. According to Liu in

consuming for some tasks [13]. In order to produce

the sentiment score, Hu Lui and Vader's methods are

mostly applied in sentiment analysis [14][15].

to identify the COVID-19 pandemic in Twitter posts

using standard machine learning-based techniques

such as SVM, Naive Bayes, Logistic Regression, and

others, with the aid of the term frequency-inverse

document frequency (TF-IDF), [19]. The results of

the experiments show that the proposed 24 approach

is promising in detecting the COVID-19 pandemic in

Twitter messages, with overall accuracy, precision,

recall, and F1 score between 70% and 80% and the

confusion matrix for machine learning approaches

complexity. Some studies had performed the

sentiment analysis in airlines data set, such as Gupta

and Bhargav [20] and Li et al, uses the Kaggle

datasets with BERT and variants [21]. The success of

these studies had given rise to the motivation to

proceed with local airlines and come out with

meaningful analysis. Table 1 shows the

Filtering is a form of cleaning noise from raw data.

Data such as URL links, retweet counts, usernames,

hashtags, etc., are removed from the dataset. Letters

are also standardized by converting them to

lowercase. The final steps in filtering the tweets are

the removal of punctuation marks and special

characters.

Before further pre-processing techniques are carried

Stop words are words that contribute nothing to the

meaning of a sentence. Therefore, they can be safely

omitted without jeopardizing the sense of the

sentence. Examples are the, a, he, she, has, have, etc.

We will improve the model's performance by

removing the meaningless term from the model's

evaluation.

Lemmatisation is a method of converting words into

their basic forms, considering the context of thought.

labelled as positives, True Negative (TN) are

negative subjects that have been correctly labelled as

negative, and False Negative (FN) are positive

subjects that have been incorrectly labelled as

negative. Next, the accuracy of the classifiers is

calculated based on the formula:

The development of a dashboard follows the

modelling phase to visualise the sentiment analysis

results by selecting appropriate charts and graphs to

make the results more meaningful and easier to

interpret. The dashboard is developed using a

After the dataset has been cleaned, RapidMiner is

used to determine the polarity score of the sentiments.

This process identifies whether the tweets are

positive, negative, or neutral. This labelled dataset is

needed for later data modelling. VADER and

SentiWordNet are the models used to determine the

sentiment polarity.

Table 2 shows the total number of sentiments

stability, and each label is almost balanced, while the

SentiWordNet approach produces unbalanced total

74.10% with an average of 72.92% and an F1 score

of 73.49% with an average of 72.14% using the

VADER approach and a split of 90:10 percent. Figure

3 below shows an example of the dashboard that was

created. Figure 3 below shows an example of the

dashboard created.

[19] Amin, S., Uddin, M. I., Al-Baity, H. H., Zeb,

[20] Gupta, N., Bhargav, R. (2023). Sentiment

[21] Li, Zehong, Chuyang Yang, and Chenyu