Achmad Benny Mutiara
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Achmad Benny Mutiara has a total of 96 co-authors. Here are the top co-authors by number of shared publications:
- Rina Refianti
Gunadarma University
11 shared publications Adang Suhendra
Department of Informatics Engineering, Gunadarma University, Jakarta, Indonesia
7 shared publicationsErnastuti
Science and Information Technology, Gunadarma University, Departement of Computer
4 shared publicationsMohammad Iqbal
Faculty of Computer Science and Information Technology, Gunadarma University, Jakarta, Indonesia
4 shared publicationsAsep Juarna
Faculty of Computer Science and Information, Technology Gunadarma University, Jl. Margonda Raya No.100, Depok, 16424, Indonesia
3 shared publications- Tubagus Maulana Kusuma
Gunadarma University
3 shared publications
Latest Publications
- Open Access
- 1 August 2026
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 15 (4) , 3365-3375
WVisionBERT-VL: a multimodal model architecture for toxicity classification on social media platforms using large language models
The increasing prevalence of toxic content on social media, conveyed through text, images, and videos, poses significant challenges for automated content moderation systems. Although prior studies have reported promising results in unimodal and bimodal settings, they often fail to capture implicit and contextual toxicity emerging from interactions across multiple modalities, particularly in non-English environments. This paper proposes WVisionBERT-VL, an end-to-end multimodal framework for toxicity detection that integrates text, image, and video modalities within a unified architecture. The proposed model incorporates modality-specific encoders, bidirectional multi-head cross attention (BMHCA) for cross-modal synchronization, and an adaptive fusion gate to dynamically balance modality contributions. A balanced multimodal dataset is constructed from social media platforms, including X, Instagram, and TikTok, and refined using a model-based labeling strategy with limited human-in-the-loop validation. Experimental results on a custom Indonesian dataset demonstrate strong in-domain performance, achieving an accuracy of 94.12%, a macro-F1 of 0.9407, and a receiver operating characteristic - area under the curve (ROC-AUC) of 0.9721, with robustness further validated through five-fold cross-validation. Cross-dataset evaluation highlights challenges related to domain shift, underscoring the need for future research on robust and domain-adaptive multimodal toxicity detection.
- 1 August 2026
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 15 (4) , 3365-3375
Open AccessWVisionBERT-VL: a multimodal model architecture for toxicity classification on social media platforms using large language models
The increasing prevalence of toxic content on social media, conveyed through text, images, and videos, poses significant challenges for automated content moderation systems. Although prior studies have reported promising results in unimodal and bimodal settings, they often fail to capture implicit and contextual toxicity emerging from interactions across multiple modalities, particularly in non-English environments. This paper proposes WVisionBERT-VL, an end-to-end multimodal framework for toxicity detection that integrates text, image, and video modalities within a unified architecture. The proposed model incorporates modality-specific encoders, bidirectional multi-head cross attention (BMHCA) for cross-modal synchronization, and an adaptive fusion gate to dynamically balance modality contributions. A balanced multimodal dataset is constructed from social media platforms, including X, Instagram, and TikTok, and refined using a model-based labeling strategy with limited human-in-the-loop validation. Experimental results on a custom Indonesian dataset demonstrate strong in-domain performance, achieving an accuracy of 94.12%, a macro-F1 of 0.9407, and a receiver operating characteristic - area under the curve (ROC-AUC) of 0.9721, with robustness further validated through five-fold cross-validation. Cross-dataset evaluation highlights challenges related to domain shift, underscoring the need for future research on robust and domain-adaptive multimodal toxicity detection.
- Open Access
- 5 October 2025
- Published by Pandawan in International Journal of Cyber and IT Service Management
- Vol. 6 (1) , 65-77
Improving Smear-Negative Tuberculosis Detection Using Data Augmentation and Faster R-CNN
Smear-Negative Pulmonary Tuberculosis (SNPT) remains a major diagnostic challenge due to the low bacterial load that frequently causes false negative results in sputum microscopy. Chest X-ray imaging is commonly used as a complementary diagnostic tool however its interpretation relies heavily on expert radiologists and is prone to subjectivity. Recent developments in deep learning particularly object detection models provide promising opportunities to improve diagnostic accuracy. This study aims to develop and evaluate a deep learning based approach for SNPT detection in chest X-ray images using a Faster R-CNN model with a ResNet architecture. The proposed method applies data augmentation techniques including flipping rotation scaling and random brightness adjustment to enhance training data diversity and reduce overfitting. The model was implemented using PyTorch and evaluated using accuracy precision recall and F1 score. Experimental results indicate that data augmentation substantially improves performance achieving 76.60% accuracy and 68.57% F1 score compared to 53.06% accuracy and 51.06% F1 score without augmentation. Improved recall reflects higher sensitivity in detecting SNPT cases. These findings indicate that data augmentation enhances the robustness and generalization of Faster R-CNN models for SNPT detection and supports the potential of AI assisted diagnostic systems in tuberculosis screening programs.
- 5 October 2025
- Published by Pandawan in International Journal of Cyber and IT Service Management
- Vol. 6 (1) , 65-77
Open AccessImproving Smear-Negative Tuberculosis Detection Using Data Augmentation and Faster R-CNN
Smear-Negative Pulmonary Tuberculosis (SNPT) remains a major diagnostic challenge due to the low bacterial load that frequently causes false negative results in sputum microscopy. Chest X-ray imaging is commonly used as a complementary diagnostic tool however its interpretation relies heavily on expert radiologists and is prone to subjectivity. Recent developments in deep learning particularly object detection models provide promising opportunities to improve diagnostic accuracy. This study aims to develop and evaluate a deep learning based approach for SNPT detection in chest X-ray images using a Faster R-CNN model with a ResNet architecture. The proposed method applies data augmentation techniques including flipping rotation scaling and random brightness adjustment to enhance training data diversity and reduce overfitting. The model was implemented using PyTorch and evaluated using accuracy precision recall and F1 score. Experimental results indicate that data augmentation substantially improves performance achieving 76.60% accuracy and 68.57% F1 score compared to 53.06% accuracy and 51.06% F1 score without augmentation. Improved recall reflects higher sensitivity in detecting SNPT cases. These findings indicate that data augmentation enhances the robustness and generalization of Faster R-CNN models for SNPT detection and supports the potential of AI assisted diagnostic systems in tuberculosis screening programs.
- Open Access
- 1 December 2024
- Published by Bright Publisher in Journal of Applied Data Sciences
- Vol. 5 (4) , 2054-2066
Data Visualization of Climate Patterns in Indonesia Using Python and Looker Studio Dashboard: A Visual Data Mining Approach
Data Visualization of Climate Patterns in Indonesia Using Python and Looker Studio Dashboard: A Visual Data Mining Approach
- 1 December 2024
- Published by Bright Publisher in Journal of Applied Data Sciences
- Vol. 5 (4) , 2054-2066
Open AccessData Visualization of Climate Patterns in Indonesia Using Python and Looker Studio Dashboard: A Visual Data Mining Approach
Data Visualization of Climate Patterns in Indonesia Using Python and Looker Studio Dashboard: A Visual Data Mining Approach
- Open Access
- 24 July 2024
- Published by Pandawan in Aptisi Transactions on Technopreneurship (ATT)
- Vol. 6 (3) , 509−521
Data Governance in Blockchain-Based Systems for Internship Grade Conversion
Higher education institutions facechallenges in collaborating with the corporate sector and other academic entities, necessitating secure and effective communication channels. The Merdeka Learning program provides students with an expansive platform to acquire scientific knowledge and ensures equitable competencies among graduates. This study aims to address the increasing importance of data governance in such collaborations by proposing a blockchain-based framework specifically for internship grade conversion. Blockchain technology offers a secure and tamper-proof solution for managing and sharing data, addressing challenges in transparency and consistency. The research adopts an experimental approach combined with a comprehensive review of blockchain technology and its integration into existing data governance frameworks. The proposed outcome includes a detailed data governance framework with a prototype data management layer built on blockchain, ensuring robust data security and reliable internship evaluations. By leveraging blockchain immutability and transparency, the framework enhances the reliability of grade conversion, reduces manual errors, and promotes fair assessments. This research contributes to improving the operational efficiency of academic institutions and fostering trust among stakeholders. Furthermore, the framework aligns with the goals of Merdeka Learning by supporting seamless internship management and ensuring transparent evaluation processes. The findings aim to provide a scalable and impactful solution for higher education institutions, facilitating a more reliable and innovative approach to managing data in collaborative environments.
- 24 July 2024
- Published by Pandawan in Aptisi Transactions on Technopreneurship (ATT)
- Vol. 6 (3) , 509−521
Open AccessData Governance in Blockchain-Based Systems for Internship Grade Conversion
Higher education institutions facechallenges in collaborating with the corporate sector and other academic entities, necessitating secure and effective communication channels. The Merdeka Learning program provides students with an expansive platform to acquire scientific knowledge and ensures equitable competencies among graduates. This study aims to address the increasing importance of data governance in such collaborations by proposing a blockchain-based framework specifically for internship grade conversion. Blockchain technology offers a secure and tamper-proof solution for managing and sharing data, addressing challenges in transparency and consistency. The research adopts an experimental approach combined with a comprehensive review of blockchain technology and its integration into existing data governance frameworks. The proposed outcome includes a detailed data governance framework with a prototype data management layer built on blockchain, ensuring robust data security and reliable internship evaluations. By leveraging blockchain immutability and transparency, the framework enhances the reliability of grade conversion, reduces manual errors, and promotes fair assessments. This research contributes to improving the operational efficiency of academic institutions and fostering trust among stakeholders. Furthermore, the framework aligns with the goals of Merdeka Learning by supporting seamless internship management and ensuring transparent evaluation processes. The findings aim to provide a scalable and impactful solution for higher education institutions, facilitating a more reliable and innovative approach to managing data in collaborative environments.
- Open Access
- 24 July 2024
- Published by Pandawan in Aptisi Transactions on Technopreneurship (ATT)
- Vol. 6 (3) , 339−356
Artificial Intelligence Model for Detecting Tax Evasion Involving Complex Network Schemes
Tax evasion through complex network schemes poses a significant challenge to tax authorities, leading to substantial revenue losses. This paper aims to develop and evaluate an artificial intelligence model designed to detect tax evasion within complex corporate networks, providing a comprehensive overview and prediction of tax avoidance behaviors. Employing a systematic literature review and document analysis of applicable tax regulations, the study utilizes Social Network Analysis (SNA) as a primary technique for mapping and analyzing taxpayer networks. The process involves matching taxable identities, constructing taxpayer graphs, extracting features, and developing a machine learning model. The proposed architectures and processes demonstrate the potential for tax authorities to enhance their capabilities in detecting tax evasion involving complex networks, with the machine learning model effectively identifying features related to both individual and network characteristics of taxpayers. The findings suggest that the integration of artificial intelligence and big data analytics can significantly improve the detection of tax evasion in complex corporate structures, offering valuable tools for tax authorities to better enforce tax compliance.
- 24 July 2024
- Published by Pandawan in Aptisi Transactions on Technopreneurship (ATT)
- Vol. 6 (3) , 339−356
Open AccessArtificial Intelligence Model for Detecting Tax Evasion Involving Complex Network Schemes
Tax evasion through complex network schemes poses a significant challenge to tax authorities, leading to substantial revenue losses. This paper aims to develop and evaluate an artificial intelligence model designed to detect tax evasion within complex corporate networks, providing a comprehensive overview and prediction of tax avoidance behaviors. Employing a systematic literature review and document analysis of applicable tax regulations, the study utilizes Social Network Analysis (SNA) as a primary technique for mapping and analyzing taxpayer networks. The process involves matching taxable identities, constructing taxpayer graphs, extracting features, and developing a machine learning model. The proposed architectures and processes demonstrate the potential for tax authorities to enhance their capabilities in detecting tax evasion involving complex networks, with the machine learning model effectively identifying features related to both individual and network characteristics of taxpayers. The findings suggest that the integration of artificial intelligence and big data analytics can significantly improve the detection of tax evasion in complex corporate structures, offering valuable tools for tax authorities to better enforce tax compliance.
- Open Access
- 29 January 2024
- Published by Bright Publisher in Journal of Applied Data Sciences
- Vol. 5 (1) , 146-157
A Lexicon-Based Long Short-Term Memory (LSTM) Model for Sentiment Analysis to Classify Halodoc Application Reviews on Google Playstore
The development of information and communication technology is developing very quickly, has made many new breakthroughs. One of these technological advances is in the health sector, the creation of telemedicine applications. During the Covid-19 pandemic, it is difficult for people to get access to health. Therefore, telemedicine applications are needed. Halodoc is one of the telemedicine applications that has successfully become the top health application on the Google PlayStore. The application has been used by more than ten million users throughout Indonesia and received a rating of 4.6. To be able to see ratings and satisfaction from the public, user reviews are needed. The very large number of reviews often contain errors, making them difficult to decipher. Based on this, this research aims to create a web application, which can classify user reviews of the Halodoc application, using a proposed lexicon-based Long Short-Term Memory (LSTM) Model. Application is built using the Flask framework and the Python programming language. Models are created and trained using the TensorFlow library. The results of the model evaluation get an accuracy of 85.3% with an average precision value of 85.3%, a recall value of 85.6% and an f1-score of 85.3%. The proposed LSTM model can be used to classify Halodoc review sentiment classes.
- 29 January 2024
- Published by Bright Publisher in Journal of Applied Data Sciences
- Vol. 5 (1) , 146-157
Open AccessA Lexicon-Based Long Short-Term Memory (LSTM) Model for Sentiment Analysis to Classify Halodoc Application Reviews on Google Playstore
The development of information and communication technology is developing very quickly, has made many new breakthroughs. One of these technological advances is in the health sector, the creation of telemedicine applications. During the Covid-19 pandemic, it is difficult for people to get access to health. Therefore, telemedicine applications are needed. Halodoc is one of the telemedicine applications that has successfully become the top health application on the Google PlayStore. The application has been used by more than ten million users throughout Indonesia and received a rating of 4.6. To be able to see ratings and satisfaction from the public, user reviews are needed. The very large number of reviews often contain errors, making them difficult to decipher. Based on this, this research aims to create a web application, which can classify user reviews of the Halodoc application, using a proposed lexicon-based Long Short-Term Memory (LSTM) Model. Application is built using the Flask framework and the Python programming language. Models are created and trained using the TensorFlow library. The results of the model evaluation get an accuracy of 85.3% with an average precision value of 85.3%, a recall value of 85.6% and an f1-score of 85.3%. The proposed LSTM model can be used to classify Halodoc review sentiment classes.
- Open Access
- 7 June 2023
- Published by Asosiasi Dosen Muda Indonesia in International Journal Multidisciplinary Science
- Vol. 2 (2) , 26-33
CONCEPTUAL REGIONAL ORIGIN RECOGNITION USING CNN CONVOUTION NEURAL NETWORK ON BANDUNG, BOGOR AND CIREBON REGIONAL ACCENTS
Sound detection is a challenge in machine learning due to the noisy nature of signals, and the small amount of (labeled) data that is usually available. The need for sound detection in Indonesia is quite important because there are many community organizations that form groups according to the land of their origin. Especially in big cities, where people from various tribes gather and exchange cultures. However, it has a disadvantage that affects these tribes, namely the loss of the original culture of certain areas. The Sundanese are the object of this research, including Bandung, Bogor and Cirebon. Voice data is divided into 2 types, namely male and female, each region consists of 50 respondents with 25 male and female voices with a maximum voting time of 1 minute. The method used is CNN architecture based on supervised learning, preprocessing using MFCC (Mel Frequency Cepstral Coefficients) to obtain feature extraction from voice data. CNN architecture is carried out 3 times convolution with max pooling and dropout on each convolution.
- 7 June 2023
- Published by Asosiasi Dosen Muda Indonesia in International Journal Multidisciplinary Science
- Vol. 2 (2) , 26-33
Open AccessCONCEPTUAL REGIONAL ORIGIN RECOGNITION USING CNN CONVOUTION NEURAL NETWORK ON BANDUNG, BOGOR AND CIREBON REGIONAL ACCENTS
Sound detection is a challenge in machine learning due to the noisy nature of signals, and the small amount of (labeled) data that is usually available. The need for sound detection in Indonesia is quite important because there are many community organizations that form groups according to the land of their origin. Especially in big cities, where people from various tribes gather and exchange cultures. However, it has a disadvantage that affects these tribes, namely the loss of the original culture of certain areas. The Sundanese are the object of this research, including Bandung, Bogor and Cirebon. Voice data is divided into 2 types, namely male and female, each region consists of 50 respondents with 25 male and female voices with a maximum voting time of 1 minute. The method used is CNN architecture based on supervised learning, preprocessing using MFCC (Mel Frequency Cepstral Coefficients) to obtain feature extraction from voice data. CNN architecture is carried out 3 times convolution with max pooling and dropout on each convolution.
- Open Access
- 1 February 2022
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 12 (1) , 1018-1029
Automated hierarchical classification of scanned documents using convolutional neural network and regular expression
This research proposed automated hierarchical classification of scanned documents with characteristics content that have unstructured text and special patterns (specific and short strings) using convolutional neural network (CNN) and regular expression method (REM). The research data using digital correspondence documents with format PDF images from pusat data teknologi dan informasi (technology and information data center). The document hierarchy covers type of letter, type of manuscript letter, origin of letter and subject of letter. The research method consists of preprocessing, classification, and storage to database. Preprocessing covers extraction using Tesseract optical character recognition (OCR) and formation of word document vector with Word2Vec. Hierarchical classification uses CNN to classify 5 types of letters and regular expression to classify 4 types of manuscript letter, 15 origins of letter and 25 subjects of letter. The classified documents are stored in the Hive database in Hadoop big data architecture. The amount of data used is 5200 documents, consisting of 4000 for training, 1000 for testing and 200 for classification prediction documents. The trial result of 200 new documents is 188 documents correctly classified and 12 documents incorrectly classified. The accuracy of automated hierarchical classification is 94%. Next, the search of classified scanned documents based on content can be developed.
- 1 February 2022
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 12 (1) , 1018-1029
Open AccessAutomated hierarchical classification of scanned documents using convolutional neural network and regular expression
This research proposed automated hierarchical classification of scanned documents with characteristics content that have unstructured text and special patterns (specific and short strings) using convolutional neural network (CNN) and regular expression method (REM). The research data using digital correspondence documents with format PDF images from pusat data teknologi dan informasi (technology and information data center). The document hierarchy covers type of letter, type of manuscript letter, origin of letter and subject of letter. The research method consists of preprocessing, classification, and storage to database. Preprocessing covers extraction using Tesseract optical character recognition (OCR) and formation of word document vector with Word2Vec. Hierarchical classification uses CNN to classify 5 types of letters and regular expression to classify 4 types of manuscript letter, 15 origins of letter and 25 subjects of letter. The classified documents are stored in the Hive database in Hadoop big data architecture. The amount of data used is 5200 documents, consisting of 4000 for training, 1000 for testing and 200 for classification prediction documents. The trial result of 200 new documents is 188 documents correctly classified and 12 documents incorrectly classified. The accuracy of automated hierarchical classification is 94%. Next, the search of classified scanned documents based on content can be developed.
- 3 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
The Crowdsourcing Method to Normalize “Bahasa Alay”, a Case of Indonesian Corpus
In verbal communication, people use sentences that can be classified into two categories, namely formal and non- formal. The former meets the grammatical standard as prescribed by linguistic rules of the language, while the latter deviates it. In daily communication, however, non-formal sentences are more intensively used because they are more practical and easier to understand. With this deviation, nonformal sentences cause problems in linguistic computation because most linguistic computations use formal languages that already have standard rules. This research aims to develop an Indonesian closed corpus related to airline ticket reservations. The data used to develop the corpus are taken from conversations between customer service staff and consumers in airline ticket reservations. This is a preliminary study to propose and develop a chatbot in airline ticket reservations. The result of this study is the Indonesian closed corpus related to airline ticket reservations to determine the right response for consumers.
- 3 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
The Crowdsourcing Method to Normalize “Bahasa Alay”, a Case of Indonesian Corpus
In verbal communication, people use sentences that can be classified into two categories, namely formal and non- formal. The former meets the grammatical standard as prescribed by linguistic rules of the language, while the latter deviates it. In daily communication, however, non-formal sentences are more intensively used because they are more practical and easier to understand. With this deviation, nonformal sentences cause problems in linguistic computation because most linguistic computations use formal languages that already have standard rules. This research aims to develop an Indonesian closed corpus related to airline ticket reservations. The data used to develop the corpus are taken from conversations between customer service staff and consumers in airline ticket reservations. This is a preliminary study to propose and develop a chatbot in airline ticket reservations. The result of this study is the Indonesian closed corpus related to airline ticket reservations to determine the right response for consumers.
- Open Access
- 5 April 2020
- Published by Universitas Indraprasta PGRI in STRING (Satuan Tulisan Riset dan Inovasi Teknologi)
- Vol. 4 (3) , 330-338
Pengujian Algoritma Clustering Affinity Propagation dan Adaptive Affinity Propagation terhadap IPK dan Jarak Rumah
Clustering which is a method to classify data easily is used for a purpose of looking at the correlation among data attributes. Clustering is also a data point grouping based on similarity value to determine the cluster center. Affinity Propagation (AP) and Adaptive Affinity Propagation (Adaptive AP) are clustering algorithms that produce number of cluster, cluster members and exemplar of each cluster. This research is conducted to find out a more effective algorithm when clustering data. Besides, to know the correction offered by Adaptive AP Algorithm which is the developed form of AP Algorithm, the researcher implemented and tested both algorithms by using Matlab R2013a 8.10 with 250 data taken from students’ GPA and the distance from their houses to campus. The analysis of test result application from both algorithms shows that the best algorithm is Adaptive AP because it produces optimal clustering. Another result is no correlation between GPA and home distance.
- 5 April 2020
- Published by Universitas Indraprasta PGRI in STRING (Satuan Tulisan Riset dan Inovasi Teknologi)
- Vol. 4 (3) , 330-338
Open AccessPengujian Algoritma Clustering Affinity Propagation dan Adaptive Affinity Propagation terhadap IPK dan Jarak Rumah
Clustering which is a method to classify data easily is used for a purpose of looking at the correlation among data attributes. Clustering is also a data point grouping based on similarity value to determine the cluster center. Affinity Propagation (AP) and Adaptive Affinity Propagation (Adaptive AP) are clustering algorithms that produce number of cluster, cluster members and exemplar of each cluster. This research is conducted to find out a more effective algorithm when clustering data. Besides, to know the correction offered by Adaptive AP Algorithm which is the developed form of AP Algorithm, the researcher implemented and tested both algorithms by using Matlab R2013a 8.10 with 250 data taken from students’ GPA and the distance from their houses to campus. The analysis of test result application from both algorithms shows that the best algorithm is Adaptive AP because it produces optimal clustering. Another result is no correlation between GPA and home distance.