Corresponding Author: Abdul Robi Padri
Email: [email protected]
INTRODUCTION
Traffic jams often occur in various and The increase
in population growth in Indonesia every year has shown that Indonesia is
experiencing a population surge; this is also the background of congestion (Janah et al., 2016). The ratio of the number of private vehicles and
public transportation has also increased so that it can cause traffic jams
which take many victims of accidents in Indonesia (Pratiwi, 2016). �With the
presence of Transportation Technology innovations that make it easier for
people to travel to other places, such as going to work and school, shopping as
well as traveling, around 2011 Indonesia, a trend emerged for online public
transportation, such as online taxis and online motorcycle taxis (Gojek,
Ubermotor, Blu-jek) which in fact has helped the problems of some people
throughout Indonesia when traveling (Hariansyah, 2018). However, traffic jams in Indonesian territory have
not been optimally resolved. In recent years, Deadlock has frequently used
naive Bayes Google Collab Algorithm, then analysis. To support the research,
the author made a survey of 70 respondents about traffic in various traffic in
Indonesia. Starting from the aim of implementing online school transportation
or vehicles that the local government has provided is to reduce traffic jams,
reduce accidents, reduce tardiness for both students and students, and minimize
costs. We hope that the results of implementing public school transportation
can provide the best solutions and make useful contributions to all West Java
people in avoiding traffic jams and being more efficient, safe, and comfortable.
This is especially true in big cities due to the
increasing number of vehicles every year and the number of inadequate roads (Rasyid et al., 2020). Problems are also increasingly complex with the
increasing number of accidents that occur (Desmira et al., 2015). Transportation is the process of distributing goods to several places-the
high demand for transportation results in a lack of performance efficiency in
serving the community or customers (Ardini & Lutfiyana, 2018). So we need a Transportation
Method to complete and optimize Transportation performance to save costs and
distances (Hermanto et al., 2017). Based on the discussion of
these conditions, the authors try to offer a strategy for dealing with traffic
jams in West Java with the concept of Business Intelligence. This article offers a model for implementing Business
Intelligence-based online school transportation using the Naive Bayes algorithm
method. The dataset is a recap of West Java road traffic jam data daily on
Google Maps 2017, accessed online; this dataset is tested for accuracy (Darmawan & Makruf,
2023).
Based on the background above, this research aims to
identify and analyze developing applications using the Naive Bayes
classification method. Applications that will be developed in the future will
classify tweet data that contains traffic information in Indonesia. After the
data is classified, then data visualization is carried out (Dewa et al., 2021). The benefit of this research is that it can help
develop applications that can accurately classify data. In the context of
traffic congestion in Indonesia, this application can help users to obtain
relevant information about real-time traffic conditions. This information can
be used to make informed decisions and help reduce traffic congestion problems
in Indonesia (Habiba et al., 2023). In addition, application development using the Naive
Bayes method can be applied to various fields, such as email spam detection and
text classification on social media.
METHODS
Research Stage
The stages of this research were carried out in 8 stages,
each stage from obtaining data to testing the Naive Bayes method and the linear
model and RandomForest. The steps taken.
Figure 1. Trending Topic
Analysis Flowchart
1. Data collection is a Twitter website, and by registering
as a developer account, you will get an API where this API can be used for
crawling Twitter data using orange software and by searching for certain topic
keywords and producing the amount of data obtained is 1501 (Kasogi et al., 2020).
2. It is determining the Topics to be labeled in the
analysis of the preliminary dataset, which is from the relevant and not
relevant preliminary labels as well as the media or individuals that will be
obtained from the manual labeling.
3. Preprocessing
Preprocessing aims to change unstructured data into
structured data because it will produce data that can be easily processed
according to their needs (Harjanta, 2015). The first preprocessing process is deleting retweet
data from tweets because retweets cannot explain the personality of the Twitter
account user. They were then followed by deleting (URL, RT, HTTP, #, @,
removing hashtags, removing character spaces from left and right text, and
replacing newline into the space). Case folding is changing all the characters
in the text to lowercase. Tokenization or string splitting, the text becomes a
token list, then the tweet data is broken down into word units. Filtering is
the selection of important words after the tokenizing process. Moreover, the
final stage is steamer-reducing words and changing words affixed to basic
words.
4. Polarity labeling, where the polarity score determine the
polarity of Tweet Sentiment with the Indonesian Sentiment Lexicon Loading
positive and negative and neutral data lexicons.
5. TF-IDF weighting
In this weighting, each Twitter dataset that has been
preprocessed will be given a weighting value using TF-IDF which will produce a
value for each text dataset from the tweet (Wahyunita et al.,
2020).
6. Na�ve Bayes Algorithm Classification
The data that has been preprocessed continues to enter
the classification stage using the Na�ve Bayes algorithm and RandomForest as a
classifier.
7. Prediction Models
At this stage, the data after all the preprocessing and
processing with algotima using na�ve Bayes RandomFores and using
CountVectorizer only with trigrams or using TF-IDF vectors for text
preprocessing. The prediction model will produce predictive results from the
classification of the Algorithm and the data that has been tested, which has
been output to an accurate accuracy value.
8. performance evaluation
In this process, accuracy, precision, and Recall
calculations will be carried out to measure the system that has been made.
Software and Hardware Requirements
Analysis
At this stage, a software and hardware requirements analysis is needed
to run a project researchers are developing using needed system design in this
study (Hariansyah,
2018).
Table 1.
Software Requirements
|
Software
|
Version
|
|
Operating
system
|
Mac OS
Monterey
|
|
Application
1
|
Orange3
|
|
Application
2
|
Google Colab
|
Table 2. Software Requirements
|
Software
|
Version
|
|
Processor
|
Intel(R) Core(TM) i5
|
|
Storage
(RAM)
|
16GB DDR4
|
|
HDD/SSD
|
1TB
|
RESULTS AND DISCUSSION
The research results determine the fastest route, whether to continue or look for another way. Preliminary dataset analysis
aims to obtain the results of the dataset with topics that have been used for
research methods by conducting direct manual research of these topics and
sample tables that have been adapted to the research topic dataset percentage
of relevant tweets in the datasets.
Labeling
Dataset
Based on the results of crawling data taken
from Twitter, with a total of 1501 tweets on traffic jams,
traffic jams, and traffic jams throughout Indonesia. Moreover, after preprocessing, the data will produce 1182 tweets.
Moreover, where the data has resulted in polarity score labeling with the
Indonesian lexicon dictionary, which gets an accuracy value by obtaining
sentiment analysis labeling, Negative, Positive, and Neutral produce the
analysis data automatically with the Indonesian dictionary. Moreover, do the
preliminary plebeian manually from the relevant and the media or individuals (Febriyani & Februariyanti,
2023).
Figure
1. The dataset that has preprocessed the trending topic analysis flow
Table 3. Polarity Percentage of
sentiment analysis Negative Neutral Positive
|
Type
|
Datasets
|
Negative
|
Neutral
|
Positive
|
|
Sentiment analysis
|
1182
|
683
|
182
|
317
|
|
Percentage Each
|
57.78%
|
15.40%
|
26.82%
|
Results The total
number of sentiment analysis labels total 1182 /3,
after which will result respectively (Negative 57.8% =683 Tweets, Neutral
26.82% = 317 Tweets, Positive 15.4% = 182 Tweets).
Visual Bar Chart and Donut Chart
matplotib python
|
Figure 3. Generating Bar Visual
Chart
|
Figure 4. Visuals That Produce Pie
and Donut Charts for Displaying Labeling Sentiment Analysis
|
Preliminary
Datasets
Table 4. Percentage of Relevant
Tweets In Their Datasets
|
Type
|
Datasets
|
Relevant
|
Not Relevant
|
%Relevant
|
|
Relevant
|
1182
|
1047
|
135
|
88.58%
|
Table 5. The Proportion of Total
Tweets from Media And Individual Accounts
|
Type
|
Datasets
|
Relevant
|
Not Relevant
|
%Relevant
|
|
Media
|
1182
|
62
|
1120
|
94.75%
|
Results The total amount of the preliminary
dataset shown in Figure 5 is relevant data with total data (88.58% = 1047
tweets), while the Not relevant total data (11.42% = 135 Tweets) and
the results of the preliminary media dataset shown in Figure These 6 are
the total data for Individuals (94.75% = 1120 Tweets), and the Media (5.25% = 62
Tweets).
|
Figures
5 and 6. Visuals That Produce A Bar Chart To Display
The
Preliminary Labeling Preliminary Analysis
|
Figure 7. Visuals that
generate WordCloud top topic stopwords Example
WordCloud
with the topic 'stuck topic that has been successfully created! Moreover, the
result is that the selected topic can be seen in the image with the largest
font size (jammed), for example, if the topic I use is stuck or stuck. For
example, in our case, it has a total data record equal to 1182 Tweets recorded
with the mathematical calculation of the separation method (805 and 20%).
TF-DF bottling
Word weighting was done to assign accuracy values to certain data tweets
using classic NLP classifying by topic text data, using gram bag techniques,
n-grams, TF-IDF, etc., for text representation and applying different
classification algorithms.
Calculation of training and testing data using TF-IDF Naive Bayes
weighting
Every machine learning algorithm that is used requires a technique which
is a division of some sort. In this process, the entire dataset will be divided
[3] into two parts, namely Training and Testing. Each division will produce
each percentage, which can be 80% or 20%. However, this method will certainly
do the 80% and 20% division techniques that will be used to evaluate the
performance of the machine learning algorithm that will be used. As said above,
this processing method requires dividing a dataset into two subsets.
a.
The training
set is used to customize and train machine-learning models
b.
A set of
tests used to evaluate suitable machine learning models
This technique is an important step, and in most cases, in using the
division algorithm, we always use 80% of the dataset given to the topic to learn
from the training. The remaining 20% will be tested to see the accuracy of the
classification, to check whether going well or not. If the suggested Machine
Learning algorithm is not working properly, another classifier must be applied
to the data to be tested; for example, the image below explains the procedure
for splitting/splitting data in machine learning Tweets recorded with the
mathematical calculation of the splitting method from (805 and 20%) for the
three calcifiers, which are calculated as follows.
a.
Training
sets: 80% � 1182 = 80/100�
1182 = 0.8 � 1182 = 945 records
b.
Testing
set: 20% � 1182 = 20/100�
1182 = 0.2 � 1182 = 237 records
The
procedure for dividing datasets in machine learning:
Figure 8. Entri Data Set
Classification Process 3.4
The classification process stage is sourced from Twitter data and crawled
using orange3 software (Desmira
et al., 2015). Testing this classification model is
carried out on the developed system, and testing the accuracy of the
classification model from the results of accurate data generated using the
process of preprocessing and cleaning, which will produce data from 1501 will
produce 1182 tweets.
Figure
9. Image Starting Processing Data Cleaning and Testing
Classification is
determining a record or Classification with na�ve Bayes classifier google
collab python, which will show the classification process with google collab
and with the na�ve Bayes method where experiments or tests using the sklearn
pipeline module create a classification pipeline to generate data by one of
them using Count Vectorizer with unigrams and bigrams and using Multinomial
Na�ve Bayes, classifier prints a classification report As for the random forest
for testing the accuracy of the values that will be used to produce the classification
of this research data so that accuracy is obtained in the rankings that produce
data that has divided the Twitter data with (80% Trainset and 20% Test set) and
Testing data is what will be tested for accuracy in this software (Indriyani, 2019).
Model prediction and evaluation of metrics 3.5
This metric measurement summed up the
results as follows:
a.
accuracy
b.
Precision
c.
Recall
d.
F1 scores
Accuracy was calculated as the
total number of correct predictions over the total number of data sets (i.e.,
all correct/all). Accuracy rule is
An
example Of precision is,
An example For Recall is,
An example Of an F1 score is,
The description of the understanding of the accuracy, precision, Recall,
and f1 score model metrics is as follows; for example, a Comparison of a
classification system that has been trained to classify will give results in
predicting metrics that the total data results are 237 tweets that are to be
tested for get Negative, Neutral, Positive results.
Table confusion matrix Manually Using
Calculations with Microsoft Excel so that the calculations are more detailed
and also make verification reference materials how accurate with Python and manual
you can see the confusion matrix image below and the following example.
Figure 10. Calculation
Manual To Make Using Method Excels
Figure 11. Results From the Na�ve Bayes Test Accuracy
Examples of results from the confusion
matrix include precision, sensitivity (Recall), and F1 scores (F1-score). We
can use manual formulas to calculate these values using software such as
Microsoft Excel. Here is an explanation.
Accuracy Determination Stage from
Confusion Matrix and determine the results
= (131+1+18)/
(237)
The
average value of the results of the confusion matrix calculation that produces
macro averages and weighted averages can be explained in the following example:
Figure 12. �Total Result Of Confusion Matrix
The explanation regarding the
calculation formula presented is a formula that has been tested to assess its
accuracy. This formula includes an example of calculating the results of each
component, such as the total of negative, neutral, and positive values, which
will result in accuracy by dividing the calculation using an Excel worksheet. In
addition, this formula will also produce other values, such as macro values,
averages, and weight values calculated from the total of the entire calculation
table.
One example of a formula that produces an accuracy level using the Naive
Bayes method is based on the results of testing 237 tweet data. The tweet data
is divided into positive, neutral, and negative categories. For example, the
total test is divided into 131, 1, 5, 33, 1, 2, 44, 2, and 18. From these
numbers, the accuracy value can be calculated as follows: (131+1+18) / (237).
An example of a confusion matrix image
that has been tested with Google Colab and the example is as follows:
Figure 14. Confusion
Matrix From Data Testing
This image shows all the stages tested on data using
the manual method using an accurate Excel sheet formula. This includes
preliminary, relevant, and irrelevant data and testing involving around 20% of
the data to test accuracy through Na�ve Bayes. This test uses software such as
the Python notebook, Conda Python, or Orange, with particular emphasis on using
Python Colab from A to Z.
From the initially mixed data, the cleaning text
process in Python has removed irrelevant symbols to make the data more readable
and relevant. After this process, it was decided to test the data by dividing
it into 100%/20%, resulting in 237 data that would be used to process and
classify accuracy.
Several different methods are applied to Python, covering all aspects of
both manual and automated tests and the results of tests performed through
Python.
CONCLUSION
This study shows that the Naive Bayes method
can be used to analyze traffic jams in Indonesia by utilizing Twitter data.
Through preprocessing, sentiment analysis, and classification, researchers can identify
relevant information about traffic jams and classify them accurately. These
results can be considered for decision-making and strategic planning in
overcoming the problem of traffic.
REFERENCES
Ardini, A., & Lutfiyana, N. (2018). Metode Transportasi
Untuk Mengoptimalkan Biaya Pengiriman Barang Pada PT Trimuda Nuansa Citra
Jakarta. Information System For Educators And Professionals: Journal Of
Information System, 3(1), 55�66.
Darmawan, A., & Makruf, M. (2023). Deteksi Gaya Belajar
Siswa SMA pada Virtual Based Learning Environment (VBLE) dengan Decision Tree
C4. 5 dan Naive Bayes. KLIK: Kajian Ilmiah Informatika Dan Komputer, 3(5),
532�544. https://doi.org/10.30865/klik.v3i5.760
Desmira, D., Kautsar, A., & Darmawan, A. W. (2015). Prototype
Perancangan Informasi Kemacetan Jalan Tol Berbasis Mikrokontroler At89s52
Dengan Tampilan LCD. PROSISKO: Jurnal Pengembangan Riset Dan Observasi
Sistem Komputer, 2(2).
Dewa, W. A., Maknunah, J., & Putri, A. D. (2021).
Penerapan Metode Na�ve Bayes untuk Menentukan Pengajuan Polis Baru pada PT.�XYZ.�
Jurnal Ilmiah Komputasi, 20(1), 83�92. https://doi.org/10.32409/jikstik.20.1.2696
Febriyani, E., & Februariyanti, H. (2023). Analisis
Sentimen Terhadap Program Kampus Merdeka Menggunakan Algoritma Naive Bayes
Classifier Di Twitter. Jurnal Tekno Kompak, 17(1), 25�38. https://doi.org/10.33365/jtk.v17i1.2061
Habiba, A., Isnanto, R. R., & Suseno, J. E. (2023). The
Effect of Chi Square Feature Selection on the Na�ve Bayes Algorithm on the
Analysis of Indonesian Society�s Sentiment About Face-to-Face Learning During
the Covid-19 Pandemic. JST (Jurnal Sains Dan Teknologi), 12(1).
Hariansyah, M. (2018). Millenials �Bukan Generasi Micin.�
Guepedia.
Harjanta, A. T. J. (2015). Preprocessing Text untuk
Meminimalisir Kata yang Tidak Berarti dalam Proses Text Mining. Jurnal
Informatika Upgris, 1(1 Juni). https://doi.org/10.26877/jiu.v1i1%20Juni.804
Hermanto, N., Hermaliani, E. H., & Sutinah, E. (2017).
Vogell�s Aproximation Method dalam Optimalisasi Biaya Transportasi Pengiriman
Koran pada PT. Arah Medialog Pembangunan. Jurnal Teknik Komputer AMIK BSI,
3(1), 30�36. https://doi.org/10.31294/jtk.v3i1.1340
Indriyani, L. (2019). Analisis penerapan Na�ve Bayes untuk
memprediksi resiko kredit anggota koperasi keluarga guru. Jurnal Informatika,
6(2), 262�270. https://doi.org/10.31294/ji.v6i2.5724
Janah, S. H., Nur, S., Emil Adly, S. T., & SH, L. (2016).
Model Kebijakan Antisipatif Mengatasi Kemacetan Lalu Lintas Darat di Kota
Batam. Prosiding Seminar Nasional INDOCOMPAC.
Kasogi, I., Setiawan, E., & Syauqy, D. (2020).
Pengoptimalan Lampu Lalu Lintas menggunakan Metode Na�ve Bayes Classifier. Jurnal
Pengembangan Teknologi Informasi Dan Ilmu Komputer, 4(6), 1725�1731.
Pratiwi, R. H. (2016). Dampak Kemacetan Terhadap Kondisi
Sosial Dan Ekonomi Pengguna Jalan Di Jakarta Utara (Studi Kasus: Pegawai Kantor
Kecamatan Cilincing dan Pegawai Rumah Sakit Umum Kecamatan (RSUK) Cilincing
Jakarta Utara). Fakultas Ekonomi dan Bisnis Unpas Bandung.
Rasyid, A. D. A., Aulia, R., & Fathurrachman, M. R.
(2020). Penerapan Aplikasi Online pada Sistem Transportasi Umum Massal untuk
Meningkatkan Minat Masyarakat dalam Upaya Mengurangi Kemacetan. Sainteks,
15(2). DOI: 10.30595/sainteks.v15i2.6308
Wahyunita, S., Azhar, Y., & Hayatin, N. (2020). Analisa
Sentimen Tweet Berbahasa Indonesia dengan Menggunakan Metode Pembobotan Hybrid
TF-IDF pada Topik Transportasi Online. Jurnal Repositor, 2(2),
185�192. https://doi.org/10.22219/repositor.v2i2.238
