Oman Somantri
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
Oman Somantri has a total of 75 co-authors. Here are the top co-authors by number of shared publications:
- Linda Perdana Wanti
Politeknik Negeri Cilacap
4 shared publications Mohammad Khambali
Department of Electrical Engineering, Politeknik Negeri Semarang
3 shared publicationsMuchamad Sobri Sungkar
Department of Electrical Engineering, Politeknik Harapan Bersama Tegal, Indonesia
3 shared publications- Ratih Hafsarah Maharrani
Cilacap State Polytechnic
3 shared publications Robbi Rahim
Sekolah Tinggi Ilmu Manajemen Sukma, Medan, Indonesia
3 shared publicationsTaufiq Abidin
Department of Informatics Engineering Institut Sains & Teknologi AKPRIND Yogyakarta Yogyakarta Indonesia
3 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by Society of Visual Informatics in JOIV : International Journal on Informatics Visualization
- Vol. 10 (5) , 2038
Detecting Social Engineering Incidents from Structured Cyber Incident Data: A Rule-Based Labeling Approach with Extra Trees and Gradient Boosting Classifiers
Social Engineering (SE) remains a major cybersecurity challenge because it exploits human behavior and is difficult to identify at the incident level, particularly in structured cyber incident repositories that do not provide explicit SE labels. This study proposes a weak-labeling approach to detect SE-related incidents in structured cyber incident data, using the EuRepoC v1.3 dataset. The SE label is operationally derived from phishing-related MITRE Initial Access indicators, user-interaction requirements, and selected incident patterns, while variables used in the labeling rule are excluded from prediction to reduce direct label leakage. Two ensemble classifiers, ExtraTrees and Gradient Boosting Classifier (GBC), are evaluated under baseline and tuned settings, with class imbalance addressed using ADASYN, BorderlineSMOTE, SMOTEENN, and SMOTETomek. Performance is assessed using PR-AUC, ROC-AUC, F1-score, accuracy, precision, recall, ablation analysis, and temporal validation. Under the full feature setting, all models achieve strong performance, with PR-AUC values ranging from 0.9695 to 0.9812. The best overall result is obtained by the baseline GBC, which reaches a PR-AUC of 0.9812, ROC-AUC of 0.9782, F1-score of 0.9507, and accuracy of 0.9414. Among the Extra Trees variants, tuned ET + SMOTETomek performs most competitively. However, ablation results show a substantial drop in performance, indicating that much of the predictive power depends on impact-related variables. Temporal validation further shows that the models remain effective for newer incidents under the full-feature setting, although the derived SE label should still be interpreted with caution as a weak proxy rather than as expert-validated ground truth.
- 30 September 2026
- Published by Society of Visual Informatics in JOIV : International Journal on Informatics Visualization
- Vol. 10 (5) , 2038
Open AccessDetecting Social Engineering Incidents from Structured Cyber Incident Data: A Rule-Based Labeling Approach with Extra Trees and Gradient Boosting Classifiers
Social Engineering (SE) remains a major cybersecurity challenge because it exploits human behavior and is difficult to identify at the incident level, particularly in structured cyber incident repositories that do not provide explicit SE labels. This study proposes a weak-labeling approach to detect SE-related incidents in structured cyber incident data, using the EuRepoC v1.3 dataset. The SE label is operationally derived from phishing-related MITRE Initial Access indicators, user-interaction requirements, and selected incident patterns, while variables used in the labeling rule are excluded from prediction to reduce direct label leakage. Two ensemble classifiers, ExtraTrees and Gradient Boosting Classifier (GBC), are evaluated under baseline and tuned settings, with class imbalance addressed using ADASYN, BorderlineSMOTE, SMOTEENN, and SMOTETomek. Performance is assessed using PR-AUC, ROC-AUC, F1-score, accuracy, precision, recall, ablation analysis, and temporal validation. Under the full feature setting, all models achieve strong performance, with PR-AUC values ranging from 0.9695 to 0.9812. The best overall result is obtained by the baseline GBC, which reaches a PR-AUC of 0.9812, ROC-AUC of 0.9782, F1-score of 0.9507, and accuracy of 0.9414. Among the Extra Trees variants, tuned ET + SMOTETomek performs most competitively. However, ablation results show a substantial drop in performance, indicating that much of the predictive power depends on impact-related variables. Temporal validation further shows that the models remain effective for newer incidents under the full-feature setting, although the derived SE label should still be interpreted with caution as a weak proxy rather than as expert-validated ground truth.
- Open Access
- 30 June 2025
- Published by Politeknik Negeri Cilacap in Journal of Innovation Information Technology and Application (JINITA)
- Vol. 7 (1) , 134-142
Optimisation of Criminal Data Clustering Model using Information Gain
Crime is a phenomenon that significantly impacts society, necessitating mapping efforts that can be utilized for further analysis. Clustering, as a data analysis technique, groups objects based on similarities or differences in their characteristics. This approach enhances the understanding of data by identifying patterns and relationships between criminal events, such as crime type, time, and location. By clustering crime data based on similar characteristics, authorities can make more effective and efficient decisions in crime prevention and control. However, selecting too many attributes can negatively affect clustering performance. To address this issue, this study applies Information Gain reduction to reduce data dimensionality by eliminating attributes with low informational contribution. Additionally, three clustering methods K-Medoid, K-Means, and X-Means are compared to evaluate their performance. The concept of Information Gain is also integrated to optimize cluster formation, measuring how much an attribute contributes to distinguishing objects within a cluster. By leveraging Information Gain, this study aims to identify the most relevant and influential attributes in forming clusters that accurately represent crime data characteristics. Furthermore, the number of clusters generated is evaluated using the Davies-Bouldin Index (DBI). The results indicate that the K-Means algorithm outperforms the other two methods, achieving the best clustering quality with an optimal number of clusters (k = 6) and the lowest DBI value.
- 30 June 2025
- Published by Politeknik Negeri Cilacap in Journal of Innovation Information Technology and Application (JINITA)
- Vol. 7 (1) , 134-142
Open AccessOptimisation of Criminal Data Clustering Model using Information Gain
Crime is a phenomenon that significantly impacts society, necessitating mapping efforts that can be utilized for further analysis. Clustering, as a data analysis technique, groups objects based on similarities or differences in their characteristics. This approach enhances the understanding of data by identifying patterns and relationships between criminal events, such as crime type, time, and location. By clustering crime data based on similar characteristics, authorities can make more effective and efficient decisions in crime prevention and control. However, selecting too many attributes can negatively affect clustering performance. To address this issue, this study applies Information Gain reduction to reduce data dimensionality by eliminating attributes with low informational contribution. Additionally, three clustering methods K-Medoid, K-Means, and X-Means are compared to evaluate their performance. The concept of Information Gain is also integrated to optimize cluster formation, measuring how much an attribute contributes to distinguishing objects within a cluster. By leveraging Information Gain, this study aims to identify the most relevant and influential attributes in forming clusters that accurately represent crime data characteristics. Furthermore, the number of clusters generated is evaluated using the Davies-Bouldin Index (DBI). The results indicate that the K-Means algorithm outperforms the other two methods, achieving the best clustering quality with an optimal number of clusters (k = 6) and the lowest DBI value.
- Open Access
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 556-564
Fuzzy expert system design for detecting stunting
Stunting is a chronic nutritional problem that occurs in toddler due to lack of nutritional intake which results in impaired growth toddler. Usually, toddler who experience stunting are characterized by not increasing weight over a long period of time. Application utilization health which makes it easier for users to access information, one of which can be used to identify toddler who are stunted by selecting symptoms. The symptoms experienced by toddlers go through a system known as the system expert. In this research an expert system will be developed that is capable of early detection developmental disorders in toddlers using the Mamdani fuzzy method. The results obtained from this research are an expert system design for early detection of stunting using the Mamdani fuzzy method. The Mamdani fuzzy method was implemented to group the criteria for toddlers who fall into the stunting category or not from the initial data which is still gray because they are still unsure whether to categorize the toddler as having stunting or not. The detection accuracy rate using the Mamdani fuzzy method is 80.87% compared to expert diagnosis.
- 1 April 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (1) , 556-564
Open AccessFuzzy expert system design for detecting stunting
Stunting is a chronic nutritional problem that occurs in toddler due to lack of nutritional intake which results in impaired growth toddler. Usually, toddler who experience stunting are characterized by not increasing weight over a long period of time. Application utilization health which makes it easier for users to access information, one of which can be used to identify toddler who are stunted by selecting symptoms. The symptoms experienced by toddlers go through a system known as the system expert. In this research an expert system will be developed that is capable of early detection developmental disorders in toddlers using the Mamdani fuzzy method. The results obtained from this research are an expert system design for early detection of stunting using the Mamdani fuzzy method. The Mamdani fuzzy method was implemented to group the criteria for toddlers who fall into the stunting category or not from the initial data which is still gray because they are still unsure whether to categorize the toddler as having stunting or not. The detection accuracy rate using the Mamdani fuzzy method is 80.87% compared to expert diagnosis.
- Open Access
- 1 February 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (2) , 1169-1176
A proposed model using Naïve Bayes and generalized linear models for early detection of heart attack risk
Timely identification of diseases, particularly heart attacks is crucial for individuals, particularly the elderly, to accurately anticipate the onset of the disease based on symtoms. The objective of this study is to develop a highly accurate model for early detection of heart disease using the Naïve Bayes (NB) and generalized linear model (GLM) techniques. In addition, another concern is the model’s subfar accuracy levels, promting the implementation of measures to optimize it. The suggested approach fot optimization involves the utilization of a genetic algorithm (GA). The research findings indicate that the NB and GLM approaches achive a reasonably high level of accuracy. Specifically, the NB model achieves an accuracy of 82.53%, while the GLM achieves an accuracy of 84.50%. Following optimization, the accuracy levels notably improved, with the NB_M-GA model reaching 85.83% and the GLM_M-GA model achieving 86,48%.
- 1 February 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 33 (2) , 1169-1176
Open AccessA proposed model using Naïve Bayes and generalized linear models for early detection of heart attack risk
Timely identification of diseases, particularly heart attacks is crucial for individuals, particularly the elderly, to accurately anticipate the onset of the disease based on symtoms. The objective of this study is to develop a highly accurate model for early detection of heart disease using the Naïve Bayes (NB) and generalized linear model (GLM) techniques. In addition, another concern is the model’s subfar accuracy levels, promting the implementation of measures to optimize it. The suggested approach fot optimization involves the utilization of a genetic algorithm (GA). The research findings indicate that the NB and GLM approaches achive a reasonably high level of accuracy. Specifically, the NB model achieves an accuracy of 82.53%, while the GLM achieves an accuracy of 84.50%. Following optimization, the accuracy levels notably improved, with the NB_M-GA model reaching 85.83% and the GLM_M-GA model achieving 86,48%.
- Open Access
- 1 February 2024
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 14 (1) , 681-689
Sentiment review of coastal assessment using neural network and naïve Bayes
An assessment of a place will provide an overview for other people whether the place is feasible to be visited or not. Assessment of coastal places will provide a separate assessment for potential visitors in considering visitation. This article proposes a model using the neural network (NN) and naïve Bayes (NB) methods to classify sentiment toward coastal assessments. The proposed NN and NB models are optimized using information gain (IG) and feature weights, namely particle swarm optimization (PSO) and genetic algorithm (GA) which are carried out to increase the level of classification accuracy. Based on the experimental results, the best level of accuracy for the classification of coastal assessments is 87.11% and is named the NB IG+PSO model. The best model obtained is a model that can be used as a decision support for potential beach visitors in deciding to visit the place.
- 1 February 2024
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 14 (1) , 681-689
Open AccessSentiment review of coastal assessment using neural network and naïve Bayes
An assessment of a place will provide an overview for other people whether the place is feasible to be visited or not. Assessment of coastal places will provide a separate assessment for potential visitors in considering visitation. This article proposes a model using the neural network (NN) and naïve Bayes (NB) methods to classify sentiment toward coastal assessments. The proposed NN and NB models are optimized using information gain (IG) and feature weights, namely particle swarm optimization (PSO) and genetic algorithm (GA) which are carried out to increase the level of classification accuracy. Based on the experimental results, the best level of accuracy for the classification of coastal assessments is 87.11% and is named the NB IG+PSO model. The best model obtained is a model that can be used as a decision support for potential beach visitors in deciding to visit the place.
- Open Access
- 3 June 2023
- Published by Universitas Negeri Semarang in Scientific Journal of Informatics
- Vol. 10 (3) , 229-238
Coastal Sentiment Review Using Naïve Bayes with Feature Selection Genetic Algorithm
Purpose: The tourism potential in the maritime sector can be Indonesia's mainstay at this time, especially in enjoying the charm of the natural beauty of the coast as people know Indonesia is an archipelagic country. The purpose of this study is to find the best model by applying the feature selection genetic algorithm (GA) and Information Gain (IG) to get the best Naïve Bayes (NB) model and the best features to produce the best level of sentiment classification accuracy.Methods: The stages of the research were carried out by going through the process of searching, pre-processing, analyzing research data using the Naïve Bayes model and optimizing genetic algorithms, validating data, and model evaluation.Result: The experimental results show that the best model is naïve Bayes based on information gain and the genetic algorithm yields an accuracy rate of 86.34%.Novelty: The main contribution to this research is proposing a new model of the best NB optimization model by applying an optimization algorithm in the search for feature selection to increase sentiment classification accuracy.
- 3 June 2023
- Published by Universitas Negeri Semarang in Scientific Journal of Informatics
- Vol. 10 (3) , 229-238
Open AccessCoastal Sentiment Review Using Naïve Bayes with Feature Selection Genetic Algorithm
Purpose: The tourism potential in the maritime sector can be Indonesia's mainstay at this time, especially in enjoying the charm of the natural beauty of the coast as people know Indonesia is an archipelagic country. The purpose of this study is to find the best model by applying the feature selection genetic algorithm (GA) and Information Gain (IG) to get the best Naïve Bayes (NB) model and the best features to produce the best level of sentiment classification accuracy.Methods: The stages of the research were carried out by going through the process of searching, pre-processing, analyzing research data using the Naïve Bayes model and optimizing genetic algorithms, validating data, and model evaluation.Result: The experimental results show that the best model is naïve Bayes based on information gain and the genetic algorithm yields an accuracy rate of 86.34%.Novelty: The main contribution to this research is proposing a new model of the best NB optimization model by applying an optimization algorithm in the search for feature selection to increase sentiment classification accuracy.
- Open Access
- 22 May 2023
- Published by Indonesian Society of Applied Science (ISAS) in Journal of Applied Community Engagement
- Vol. 3 (1) , 1-9
Pelatihan Public Speaking Upaya Peningkatan Kemampuan Komunikasi Bagi Calon Duta Kampus di Politeknik Negeri Cilacap
Limitations of someone's ability to communicate in public places become an obstacle for everyone. A student, especially a campus ambassador is required to have good communication ability, therefore the information conveyed can be received by all parties. To overcome the problems faced by prospective campus ambassadors, in this article an activity is proposed in the form of public speaking training for Campus Ambassador nominees at the Cilacap State Polytechnic. The training is carried out using the lecture and role-play method. The stages of the activity are carried out through several stages, namely the stages of planning, implementing, and evaluating. The results of the training have a significant impact on the level perception of knowledge of the trainees with an average of 70% understanding of the material presented.
- 22 May 2023
- Published by Indonesian Society of Applied Science (ISAS) in Journal of Applied Community Engagement
- Vol. 3 (1) , 1-9
Open AccessPelatihan Public Speaking Upaya Peningkatan Kemampuan Komunikasi Bagi Calon Duta Kampus di Politeknik Negeri Cilacap
Limitations of someone's ability to communicate in public places become an obstacle for everyone. A student, especially a campus ambassador is required to have good communication ability, therefore the information conveyed can be received by all parties. To overcome the problems faced by prospective campus ambassadors, in this article an activity is proposed in the form of public speaking training for Campus Ambassador nominees at the Cilacap State Polytechnic. The training is carried out using the lecture and role-play method. The stages of the activity are carried out through several stages, namely the stages of planning, implementing, and evaluating. The results of the training have a significant impact on the level perception of knowledge of the trainees with an average of 70% understanding of the material presented.
- Open Access
- 8 May 2023
- Published by Universitas Gadjah Mada in Jurnal Nasional Teknik Elektro dan Teknologi Informasi
- Vol. 12 (2) , 110-116
Model Library Support Vector Machine (LibSVM) untuk Sentiment Review Penilaian Pesisir Pantai
Improving services as an effort to provide the convenience of tourist destinations, especially on the south coast of Java, is a demand placed on tourism managers, which in the long run will yield positive impacts. The assessment is conducted to determine whether the tourism destination give positive impressions to the tourists. The application of machine learning-based text mining technology, especially a sentiment review, is one of the solutions proposed to overcome this problem, therefore predictions of coastal tourism potential can be known beforehand. This research proposed a coastal sentiment review model using the library support vector machine (LibSVM) method. The process proposed a model optimization based on feature weights using the particle swarm optimization (PSO) algorithm as a model optimization to increase the accuracy level. Efforts to improve the accuracy of the proposed model are the main contribution of this research. The results of research and experiments on the proposed model produced the best model named LibSVM_IG+PSO using the information gain (IG). On the other hand, PSO-based LibSVM method generated an accuracy level of 88.97%. The model proposed in this research is expected to serve as a decision support for tourists, government, and tourism managers in assessing sentiment towards the coastal maritime tourism.
- 8 May 2023
- Published by Universitas Gadjah Mada in Jurnal Nasional Teknik Elektro dan Teknologi Informasi
- Vol. 12 (2) , 110-116
Open AccessModel Library Support Vector Machine (LibSVM) untuk Sentiment Review Penilaian Pesisir Pantai
Improving services as an effort to provide the convenience of tourist destinations, especially on the south coast of Java, is a demand placed on tourism managers, which in the long run will yield positive impacts. The assessment is conducted to determine whether the tourism destination give positive impressions to the tourists. The application of machine learning-based text mining technology, especially a sentiment review, is one of the solutions proposed to overcome this problem, therefore predictions of coastal tourism potential can be known beforehand. This research proposed a coastal sentiment review model using the library support vector machine (LibSVM) method. The process proposed a model optimization based on feature weights using the particle swarm optimization (PSO) algorithm as a model optimization to increase the accuracy level. Efforts to improve the accuracy of the proposed model are the main contribution of this research. The results of research and experiments on the proposed model produced the best model named LibSVM_IG+PSO using the information gain (IG). On the other hand, PSO-based LibSVM method generated an accuracy level of 88.97%. The model proposed in this research is expected to serve as a decision support for tourists, government, and tourism managers in assessing sentiment towards the coastal maritime tourism.
- Open Access
- 31 December 2022
- Published by Universitas Budi Darma in Jurnal Sistem Komputer dan Informatika (JSON)
- Vol. 4 (2) , 272-279
Penerapan Feature Selection Pada Algoritma Decision Tree Untuk Menentukan Pola Rekomendasi Dini Konseling
Early detection in providing recommendations for student counseling is very important, therefore you can assess the student's potential, beliefs, and attitude as early as possible. The problem that arises in this case is how to detect a student early so that he or she needs counseling assistance or not so that it can be identified early to minimize the risk of further psychological conditions. This article proposes a data mining model using a decision tree to classify counseling recommendations for students. In addition, to improve the resulting accuracy performance, a feature selection method is proposed using forward selection and genetic algorithms. The stages of the research were carried out by pre-processing the data, implementing algorithms, validating data, and optimizing the model. The experimental results show that the best level of accuracy using the decision tree model is 95.64%. It increases to 96.91% after optimization using the genetic algorithm.
- 31 December 2022
- Published by Universitas Budi Darma in Jurnal Sistem Komputer dan Informatika (JSON)
- Vol. 4 (2) , 272-279
Open AccessPenerapan Feature Selection Pada Algoritma Decision Tree Untuk Menentukan Pola Rekomendasi Dini Konseling
Early detection in providing recommendations for student counseling is very important, therefore you can assess the student's potential, beliefs, and attitude as early as possible. The problem that arises in this case is how to detect a student early so that he or she needs counseling assistance or not so that it can be identified early to minimize the risk of further psychological conditions. This article proposes a data mining model using a decision tree to classify counseling recommendations for students. In addition, to improve the resulting accuracy performance, a feature selection method is proposed using forward selection and genetic algorithms. The stages of the research were carried out by pre-processing the data, implementing algorithms, validating data, and optimizing the model. The experimental results show that the best level of accuracy using the decision tree model is 95.64%. It increases to 96.91% after optimization using the genetic algorithm.
- Open Access
- 22 April 2022
- Published by P3M Politeknik Negeri Indramayu in JTT (Jurnal Teknologi Terapan)
- Vol. 8 (1) , 17-24
MODEL SUPPORT VEKTOR MACHINE (SVM) BERDASARKAN PARAMETER WINDOWS UNTUK PREDIKSI KEKUATAN GEMPA BUMI
Earthquakes are a type of natural disaster that currently cannot be predicted. Predicting the value of earthquake magnitude for related parties such as government and National Disaster Management Authority is very important. Furthermore, the results of earthquake predictions by several parties are used as indicators in post-earthquake response in minimizing the risks that will occur. Several studies have applied machine learning methods to predict earthquakes such as deep neural networks and parallel Support Vector Regression. In this article, we propose a data mining method using the Support Vector Machine (SVM) algorithm accompanied by the optimization of the windowing parameter value in the model that is applied to predict the value of the earthquake magnitude. Based on its advantages, the SVM model was chosen because it has been applicable in time series data processing. In the experimental stage process, parameter settings are first carried out, namely setting the kernel type, sampling type, and number of windowing to optimize the level of accuracy of the resulting model. The results showed that the best model with the smallest Root Mean Square Error (RMSE) was 0.712.
- 22 April 2022
- Published by P3M Politeknik Negeri Indramayu in JTT (Jurnal Teknologi Terapan)
- Vol. 8 (1) , 17-24
Open AccessMODEL SUPPORT VEKTOR MACHINE (SVM) BERDASARKAN PARAMETER WINDOWS UNTUK PREDIKSI KEKUATAN GEMPA BUMI
Earthquakes are a type of natural disaster that currently cannot be predicted. Predicting the value of earthquake magnitude for related parties such as government and National Disaster Management Authority is very important. Furthermore, the results of earthquake predictions by several parties are used as indicators in post-earthquake response in minimizing the risks that will occur. Several studies have applied machine learning methods to predict earthquakes such as deep neural networks and parallel Support Vector Regression. In this article, we propose a data mining method using the Support Vector Machine (SVM) algorithm accompanied by the optimization of the windowing parameter value in the model that is applied to predict the value of the earthquake magnitude. Based on its advantages, the SVM model was chosen because it has been applicable in time series data processing. In the experimental stage process, parameter settings are first carried out, namely setting the kernel type, sampling type, and number of windowing to optimize the level of accuracy of the resulting model. The results showed that the best model with the smallest Root Mean Square Error (RMSE) was 0.712.