Jurnal KomtekInfo

Analysis of Strategies for Improving Learning Quality Based on Naive Bayes and Support Vector Machines

2026-01-31T15:13:32+07:00

Islamic educational institutions, particularly Islamic boarding schools, face increasing challenges in improving the quality of learning. The learning quality in Islamic boarding schools should be analyzed in depth to support effective improvement strategies. Based on this background, this study aims to classify strategies for enhancing learning quality using the Naïve Bayes and Support Vector Machine (SVM) algorithms. Naïve Bayes with a Gaussian distribution is widely recognized for its simplicity and accuracy in data classification. Meanwhile, Support Vector Machines (SVM) with a linear kernel are effective for linearly separable and high-dimensional data, enabling stable and efficient modeling in the context of data-driven analysis of learning quality in formal education. The data were collected through questionnaires distributed to 100 female students and 100 teachers. The variables examined include teacher competence, infrastructure, school management, student participation, and learning quality level. The analysis results indicate that the Naïve Bayes algorithm achieved superior performance with an accuracy of 90%, precision of 95.65%, recall of 83.33%, and an F1-score of 86.56%. In contrast, the Support Vector Machine (SVM) obtained an accuracy of 80%, precision of 58.97%, recall of 66.67%, and an F1-score of 62.32%.These findings demonstrate that Naïve Bayes provides more stable classification performance across all learning quality categories. Conversely, the Support Vector Machine (SVM) shows less optimal performance in the low-quality class due to the limited number of data samples. This study contributes effectively to the classification of learning quality levels in Islamic boarding schools

Analysis of Student Selection Models Using K-Means Clustering and K-Nearest Neighbor Classification Algorithms

2026-01-31T15:12:51+07:00

The high level of student interest in the selection process poses challenges, including student admission management. The selection process generally consists of several stages, ranging from administrative tests, academic tests, psychological tests, and physical fitness tests. Based on this, the purpose of this study is to develop an approach that can help evaluate student readiness objectively and based on data. This study aims to analyze student selection by applying the concept of data mining using the K-Means and K-Nearest Neighbor (KNN) algorithms. The K-Means algorithm is used to group student data into several clusters based on the similarity of characteristics. Meanwhile, the K-Nearest Neighbor algorithm works by classifying new data based on similarity or the closest distance. The research dataset consists of 124 student data points obtained from the Arka Padang tutoring center headquarters. Based on this study, the results show that the application of the K-Means and K-Nearest Neighbor (KNN) algorithms demonstrates that both methods are capable of processing student data to identify patterns and levels of readiness for selection, achieving an accuracy of 92.10%. Thus, this method is considered reliable in supporting the process of evaluating student readiness. This research contributes to the understanding of the application of data mining concepts to evaluate and analyze student readiness levels and demonstrates how the K-Means and KNN algorithms can be used in the process of classifying students objectively and based on data.

Automated Medical Classification of Human Brain Tumors Leveraging the Xception Convolutional Neural Network

2026-01-31T15:05:09+07:00

Student major recommendations are compiled to help students in the Communication Studies Program at the Faculty of Social Sciences, Dehasen University, Bengkulu, determine the most suitable concentration based on academic characteristics and learning data patterns. Based on this, the purpose of this study is to analyze major recommendations for students in the Communication Studies Program at the Faculty of Social Sciences, Dehasen University, Bengkulu, based on course grades.The Simple Multi-Attribute Rating Technique (SMART) was used in the process of weighting and assessing academic criteria, while K-Means was used to form major clusters based on the assessment results. This data set consists of 103 communication science students from the Faculty of Social Sciences obtained from the Dehasen Bengkulu University academic information system portal. The results of this study can recommend majors for Communication Science students at the Faculty of Social Sciences, Dehasen Bengkulu University, based on a decision support system. Based on the research results, Journalism was the most popular major with 63 students. This shows that students are more interested in journalism than in other majors. Meanwhile, Public Relations was chosen by 40 students. The contribution of this research is to improve the accuracy of student major selection in the Communication Studies Program, Faculty of Social Sciences, Dehasen University of Bengkulu. The use of the K-Means algorithm and the SMART method enables the Communication Studies Program, Faculty of Social Sciences, Dehasen University of Bengkulu to be more objective and efficient in the process of managing academic majors.

Implementation of Isolation forest for Anomaly Detection in Hospital Management Information System

2026-02-24T05:14:57+07:00

The digitization of the healthcare sector through the Hospital Management Information System (HMIS) increases the risk of patient data security due to the potential for unauthorized access and misuse of sensitive information. The large volume of activity log data makes conventional sampling-based auditing processes ineffective in identifying security threats in real time. This study aims to implement user data access variables into the Isolation Forest algorithm framework to build an intelligent anomaly detection mechanism in the information system of Dr. M. Djamil Padang General Hospital. The research methodology applies an unsupervised machine learning-based Isolation Forest algorithm to isolate deviant behavior through anomaly scoring on random isolation trees. The pre-processing stage involves extracting request, status, and data size variables and performing numerical transformation using Z-Score standardization to maintain computational stability. The research dataset is sourced from the activity logs of the HMIS web server at Dr. M. Djamil Padang General Hospital, with a total sample of 5000 user access transactions. The analysis results show that the model successfully identified 718 data points, or 14.36%, as anomalies with an accuracy rate identical to that of manual calculations. The application and implementation of the Isolation Forest technique proved to be effective in solving the problem of early detection of suspicious data traffic patterns in large data flows. The contribution of this research enhances hospital information security management through a data-based early warning system to improve the overall accountability of health information access.

Model for identifying high-achieving students using the k-means clustering algorithm and c4.5 classification

2026-01-31T15:04:50+07:00

Student achievement refers to academic accomplishments or results obtained by students in the field of education, which can influence the process of determining student academic grades, class achievement, and accomplishments. This process plays a strategic role in supporting objective educational decision-making, especially in the preparation of coaching programs, the establishment of awards, and the continuous development of student potential. Based on this, the purpose of this study is to analyze data on high-achieving students using the K-Means and C4.5 algorithms. The research methods used include K-means, which functions to group student data into different groups. C4.5 classification is capable of analyzing data characteristic similarities, and the results of the decision tree are used as the basis for the decision-making process. The dataset in this study consisted of 345 students from SMK Negeri 3 Padangsidimpuan. Based on the results of this study, it was proven that the application of the K-Means and C4.5 algorithms could achieve an accuracy of 98.59%. This research contributes to identifying high-achieving students at SMKN 3 Padangsidimpuan using the K-Means and C4.5 algorithms, which can assist the school in formulating more effective and targeted guidance policies and presenting the results of identifying high-achieving students after clustering and decision tree analysis. This serves as a basis for decision-making in determining student development programs based on objectively identified academic and non-academic achievement clusters.

Lossy Image Compression Analysis using JPEG Algorithm on Rice Leaf Disease Dataset

2026-07-13T10:41:57+07:00

The development of digital image processing technology has increased the need for more efficient image data storage and transmission, especially in the field of smart agriculture that utilizes digital images as a source of information. Rice leaf disease image datasets generally have quite large file sizes, requiring compression techniques to save storage capacity and speed up the data exchange process. This study aims to analyze the performance of lossy image compression using the JPEG algorithm on rice leaf disease datasets by evaluating file size efficiency and the quality of the compressed images. The research process includes collecting rice leaf disease image datasets, applying JPEG compression at several quality levels (quality factors), and measuring performance using the Compression Ratio (CR), Peak Signal-to-Noise Ratio (PSNR), and Mean Squared Error (MSE) parameters. The analysis results show that the lower the quality factor value used, the greater the compression ratio obtained, but the visual quality of the image decreases as indicated by the increase in the MSE value and the decrease in the PSNR value. Conversely, a higher quality factor is able to maintain image quality with the consequence of a larger file size. The findings of this study indicate that the JPEG algorithm is able to provide a good compromise between storage efficiency and visual image quality so that it remains suitable for use in rice leaf disease datasets, especially as a preprocessing stage in image processing and artificial intelligence systems for plant disease classification.

Building Expert Digital Twins through Retrieval-Augmented AI Personas: A Framework for Preserving and Transferring Human Expertise

2026-06-18T17:10:08+07:00

The preservation and transfer of human expertise represent persistent challenges in knowledge management, particularly when tacit knowledge—embedded within individual experience, judgment, and contextual reasoning—is difficult to document, scale, or disseminate. Although Large Language Models (LLMs) have enabled sophisticated conversational AI systems, existing implementations frequently exhibit factual inconsistencies, hallucinations, and inadequate representation of domain-specific expert reasoning. These deficiencies diminish the reliability of AI-mediated knowledge transfer in high-stakes educational, organizational, and professional contexts. This study addresses these limitations by proposing a framework for building Expert Digital Twins through Retrieval-Augmented AI Personas, providing a scalable and reliable mechanism for preserving and transferring human expertise. The proposed framework integrates six interconnected layers: knowledge acquisition from multimodal expert sources, preprocessing and semantic chunking, vector-based knowledge repository construction, retrieval-augmented generation, persona-driven interaction modeling, and multi-dimensional evaluation. Expert knowledge is systematically collected from books, interviews, speeches, academic articles, and digital media, then transformed into a structured semantic repository enabling dynamic, context-sensitive retrieval. Human-centered design principles are applied throughout to ensure authenticity, transparency, and user trust. An experimental evaluation was conducted by constructing an Expert Digital Twin from a domain expert's knowledge corpus and comparing its performance against a conventional LLM-based baseline using metrics including Faithfulness, Response Accuracy, Expert Similarity, Hallucination Rate, and User Trust. Results demonstrate that the retrieval-augmented AI persona substantially improves factual consistency, perceived authenticity, and knowledge transfer effectiveness. This study contributes a theoretically grounded and practically deployable framework that positions Expert Digital Twins as a novel paradigm for sustainable digital intelligence in knowledge-intensive domains.