Santi Prayudani
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Santi Prayudani has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Yuyun Yusnida Lase
Politeknik Negeri Medan
4 shared publications - Arif Ridho Lubis
Politeknik Negeri Medan
2 shared publications - Ainul Hizriadi
Universitas Sumatera Utara
1 shared publication - Al-Khowarizmi Al-Khowarizmi
Universitas Muhammadiyah Sumatera Utara
1 shared publication - Fitria Nova Hulu
Politeknik Negeri Medan
1 shared publication - Gabriel Ardi Hutagalung
Politeknik Negeri Medan
1 shared publication
Latest Publications
- Open Access
- 29 October 2026
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 5 (2 (143)) , 71-85
Assessing the opportunity of a hybrid YOLOv8- MobileNetV2 robot vision system with hyperparameter tuning for automatic organic and non-organic waste sorting
The object of this study is the automated sorting process of organic and non-organic waste using robot vision. The problem addressed is the need to automatically determine both the location and category of waste objects to support the sorting process. Waste images were processed using YOLOv8 (Ultralytics, United States) for object detection and localization and MobileNetV2 (Google, United States) for waste classification, while hyperparameter tuning was used to optimize model performance. The results showed that YOLOv8 achieved 96% for glass and paper, 94% for metal, 93% for organic waste, and 92% for cardboard. MobileNetV2 achieved classification rates of 97% for cardboard, glass, and metal, 96% for paper, 91% for plastic, and 100% for organic waste and plastic bottles. The distinctive feature of the proposed approach is the combination of object localization and classification, which provides both the position and category of waste required for automated sorting. The obtained results are explained by the complementary functions of the models: YOLOv8 localizes waste objects, while MobileNetV2 extracts their visual features for classification, with hyperparameter tuning providing a more suitable training configuration. The developed approach can be applied to camera-based robot vision and conveyor waste-sorting systems under image acquisition conditions similar to those represented in the training dataset
- 29 October 2026
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 5 (2 (143)) , 71-85
Open AccessAssessing the opportunity of a hybrid YOLOv8- MobileNetV2 robot vision system with hyperparameter tuning for automatic organic and non-organic waste sorting
The object of this study is the automated sorting process of organic and non-organic waste using robot vision. The problem addressed is the need to automatically determine both the location and category of waste objects to support the sorting process. Waste images were processed using YOLOv8 (Ultralytics, United States) for object detection and localization and MobileNetV2 (Google, United States) for waste classification, while hyperparameter tuning was used to optimize model performance. The results showed that YOLOv8 achieved 96% for glass and paper, 94% for metal, 93% for organic waste, and 92% for cardboard. MobileNetV2 achieved classification rates of 97% for cardboard, glass, and metal, 96% for paper, 91% for plastic, and 100% for organic waste and plastic bottles. The distinctive feature of the proposed approach is the combination of object localization and classification, which provides both the position and category of waste required for automated sorting. The obtained results are explained by the complementary functions of the models: YOLOv8 localizes waste objects, while MobileNetV2 extracts their visual features for classification, with hyperparameter tuning providing a more suitable training configuration. The developed approach can be applied to camera-based robot vision and conveyor waste-sorting systems under image acquisition conditions similar to those represented in the training dataset
- Open Access
- 27 June 2026
- Published by The Indonesian Institute of Science and Technology Research (IISTR) in Engineering Science Letter
- Vol. 5 (02) , 119-126
Multi-Modal Deep Learning Approach for Waste Management: Integrating Image Classification and Text Mining for Environmental Awareness
Environmental degradation caused by inefficient waste management remains a major global challenge, largely due to the limitations of conventional systems that rely on manual waste sorting and limited utilization of heterogeneous data sources. This study proposes a novel multi-modal deep learning framework that integrates visual and textual information to enhance waste classification performance while simultaneously providing insights into environmental awareness. The proposed framework combines convolutional neural networks (CNNs) for waste image classification and a recurrent neural network with long short-term memory (LSTM) architecture for text analysis. Visual and textual feature representations are integrated through a feature-level fusion strategy using vector concatenation before final classification. The image dataset consists of six waste categories, cardboard, glass, metal, paper, plastic, and trash, while the textual dataset contains waste management descriptions, community feedback, and environmental discourse collected from public and field sources. Environmental awareness was assessed through text mining by identifying dominant themes related to recycling practices, waste sorting behavior, environmental responsibility, and public concern regarding pollution and sustainability issues. Experimental results demonstrate that the proposed multimodal framework achieves an accuracy of 88.9% and an F1-score of 0.89, outperforming image-only and text-only models with accuracies of 78.4% and 81.2%, respectively. This corresponds to absolute performance improvements of 10.5% over the image-based model and 7.7% over the text-based model, while reducing the classification error rate by 40.96%. Furthermore, the multimodal model exhibits superior robustness under degraded data conditions, with only a 4.7% reduction in accuracy compared to larger performance declines observed in unimodal approaches. The main contribution of this study lies in the integration of waste image recognition and environmental-awareness extraction within a unified multimodal learning framework, enabling not only accurate waste categorization but also the generation of behavioral and sustainability-related insights that support more intelligent and sustainable waste management systems.
- 27 June 2026
- Published by The Indonesian Institute of Science and Technology Research (IISTR) in Engineering Science Letter
- Vol. 5 (02) , 119-126
Open AccessMulti-Modal Deep Learning Approach for Waste Management: Integrating Image Classification and Text Mining for Environmental Awareness
Environmental degradation caused by inefficient waste management remains a major global challenge, largely due to the limitations of conventional systems that rely on manual waste sorting and limited utilization of heterogeneous data sources. This study proposes a novel multi-modal deep learning framework that integrates visual and textual information to enhance waste classification performance while simultaneously providing insights into environmental awareness. The proposed framework combines convolutional neural networks (CNNs) for waste image classification and a recurrent neural network with long short-term memory (LSTM) architecture for text analysis. Visual and textual feature representations are integrated through a feature-level fusion strategy using vector concatenation before final classification. The image dataset consists of six waste categories, cardboard, glass, metal, paper, plastic, and trash, while the textual dataset contains waste management descriptions, community feedback, and environmental discourse collected from public and field sources. Environmental awareness was assessed through text mining by identifying dominant themes related to recycling practices, waste sorting behavior, environmental responsibility, and public concern regarding pollution and sustainability issues. Experimental results demonstrate that the proposed multimodal framework achieves an accuracy of 88.9% and an F1-score of 0.89, outperforming image-only and text-only models with accuracies of 78.4% and 81.2%, respectively. This corresponds to absolute performance improvements of 10.5% over the image-based model and 7.7% over the text-based model, while reducing the classification error rate by 40.96%. Furthermore, the multimodal model exhibits superior robustness under degraded data conditions, with only a 4.7% reduction in accuracy compared to larger performance declines observed in unimodal approaches. The main contribution of this study lies in the integration of waste image recognition and environmental-awareness extraction within a unified multimodal learning framework, enabling not only accurate waste categorization but also the generation of behavioral and sustainability-related insights that support more intelligent and sustainable waste management systems.
- Open Access
- 1 January 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (1) , 269-277
Deep neural networks approach with transfer learning to detect fake accounts social media on Twitter
The massive use of social media makes people take actions that have a negative impact on cyberspace, such as creating fake accounts that aim to commit crimes such as spam and fraud to spread false information. Fake accounts are difficult to detect in the traditional way because fake accounts always use photos, names, and unreal information, there are several criteria that can identify a fake account such as no information, few followers, and minimal activity. In the traditional model, it is difficult to detect fake accounts on many Twitters social media accounts, so the application of the deep learning model with the convolutional neural network (CNN) algorithm and the application of deep learning can help detect fake accounts. This study will use data on Twitter social media so that this research produces good accuracy for the scenarios described at the methodology stage. This research produces an accuracy of 86% for the deep learning model with the CNN algorithm, and with the traditional model, it produces an accuracy of 51% while the use of transfer learning produces an accuracy of 93.9%.
- 1 January 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (1) , 269-277
Open AccessDeep neural networks approach with transfer learning to detect fake accounts social media on Twitter
The massive use of social media makes people take actions that have a negative impact on cyberspace, such as creating fake accounts that aim to commit crimes such as spam and fraud to spread false information. Fake accounts are difficult to detect in the traditional way because fake accounts always use photos, names, and unreal information, there are several criteria that can identify a fake account such as no information, few followers, and minimal activity. In the traditional model, it is difficult to detect fake accounts on many Twitters social media accounts, so the application of the deep learning model with the convolutional neural network (CNN) algorithm and the application of deep learning can help detect fake accounts. This study will use data on Twitter social media so that this research produces good accuracy for the scenarios described at the methodology stage. This research produces an accuracy of 86% for the deep learning model with the CNN algorithm, and with the traditional model, it produces an accuracy of 51% while the use of transfer learning produces an accuracy of 93.9%.
- Open Access
- 26 January 2022
- Published by Universitas Medan Area in Journal of Informatics and Telecommunication Engineering
- Vol. 5 (2) , 238-246
Analysis of Educator Readiness in the Online Teaching Learning Process Using Naïve Bayes
This study discusses the readiness of educators in the online teaching and learning process. Samples of data were taken randomly as many as 100 (one hundred) people who were carried out using a questionnaire for educators at the junior high school level in the city of Medan. The variables used in the research are human resources, facilities and infrastructure, skills in applying technology, time management in online learning, the assessment process. Data processing and data analysis using nave Bayes algorithm. This algorithm is very well used for the process of classifying large amounts of data. The reason for using the nave Bayes algorithm in processing and analyzing data is because the way this algorithm works uses statistical and probability methods in predicting future results. The results of calculations using the nave Bayes algorithm based on the specified training data show that educators at the junior high school level are ready for the online learning process.
- 26 January 2022
- Published by Universitas Medan Area in Journal of Informatics and Telecommunication Engineering
- Vol. 5 (2) , 238-246
Open AccessAnalysis of Educator Readiness in the Online Teaching Learning Process Using Naïve Bayes
This study discusses the readiness of educators in the online teaching and learning process. Samples of data were taken randomly as many as 100 (one hundred) people who were carried out using a questionnaire for educators at the junior high school level in the city of Medan. The variables used in the research are human resources, facilities and infrastructure, skills in applying technology, time management in online learning, the assessment process. Data processing and data analysis using nave Bayes algorithm. This algorithm is very well used for the process of classifying large amounts of data. The reason for using the nave Bayes algorithm in processing and analyzing data is because the way this algorithm works uses statistical and probability methods in predicting future results. The results of calculations using the nave Bayes algorithm based on the specified training data show that educators at the junior high school level are ready for the online learning process.