Jurnal KomtekInfo

Large Language Model Method as a Translator Indonesian Into SQL Language

2025-08-26T04:50:43+07:00

The development of information technology has encouraged the massive implementation of information systems and web-based applications in various sectors, including in the academic environment. However, one of the challenges that are still often faced is the difficulty in extracting or mining information from databases flexibly without having to create additional report modules or write SQL code manually. This problem becomes an obstacle for non-technical users, such as administrative staff or lecturers, who need certain data quickly from academic information systems. In this paper, it is intended to convert Indonesian commands into SQL queries automatically, without the need to add additional programming code. Along with advances in Natural Language Processing (NLP) and Machine Learning technology with the Large Language Model (LLM) method, there is now a new approach that allows users to interact with databases only through commands in natural language. The case study was conducted on the Academic Information System of UIN Padangsidimpuan using a dataset of 1,500 student data. The focus of the research is on the type of Data Query Language (DQL) query in Indonesian form, which is then translated by the model into a SQL command to obtain the desired data. The results showed that this approach was able to achieve results with a Rouge1 conversion precision rate from 0.03 to 0.89. This shows that the integration of LLM technology in academic information systems has great potential in improving data accessibility, operational efficiency, and supporting data-driven decision-making faster and more intuitively, especially for users who do not have a technical background.

Optimization of Data Envelopment Analysis Method with MOORA in the Selection of Research Proposals and PKM

2025-08-17T14:42:29+07:00

The quality of the selection of research proposals and Community Service (PKM) of lecturers is an important element in supporting the implementation of the Tridarma of Higher Education. However, the selection process that is still carried out manually and tends to be subjective has the potential to cause bias in decision-making. This research aims to develop a decision support system that integrates the Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) and Data Envelopment Analysis (DEA) methods to increase objectivity and efficiency in the selection process of research proposals and PKM lecturers at Rokania University. The MOORA method is used to determine a preference score based on the five main criteria for research proposals and six main criteria for PKM proposals, while the DEA method is utilized to evaluate the relative efficiency of each proposal based on the ratio between the MOORA score and the amount of funding submitted. The data used in this study was obtained from the results of the assessment of three reviewers on 14 research proposals and 11 PKM proposals. Each proposal is assessed based on criteria that have been determined by LPPM, then calculations are carried out using both methods. The results show that the combination of MOORA and DEA methods is able to produce more transparent and fair proposal rankings, as well as being able to identify the most efficient proposals in the use of the budget. This study concludes that the integration of MOORA and DEA methods in the lecturer proposal selection system is able to strengthen data-based research and PKM governance, as well as make a real contribution to more rational and measurable decision-making. This system also has the potential to be further developed to support the selection of external grants, recruitment of reviewers, or the allocation of research funds nationally. These findings can be replicated in other higher education institutions that face similar challenges.

Shortest Path Navigation Optimization using Algorithm A* in GPS-Based Augmented Reality Visualization

2025-08-20T17:26:06+07:00

The development of digital technology in the Society 5.0 era encourages the integration of technology-based solutions in various fields, including higher education. Ibnu Sina University faces challenges in providing informative and accurate access to campus navigation, especially for new students and visitors. The lack of a directional system and the absence of a digital map of the campus cause difficulties in finding service locations such as the rectorate, academic room, or finance department. Conventional technologies such as Google Maps have not been able to provide specific navigation in the campus environment due to the limitations of internal building mapping. This research aims to develop a campus navigation system based on Augmented Reality (AR) and Global Positioning System (GPS) that is optimized with the A* (A-Star) algorithm for the determination of the shortest path. AR technology is used to display visual directions in real-time in a real environment through a smartphone camera. GPS plays a role in determining the user's position, while the A* algorithm calculates the shortest route based on the structure of the campus location graph using a heuristic approach. The development method used is the Multimedia Development Life Cycle (MDLC) which consists of six stages: concept, design, material collecting, assembly, testing, and distribution. Data collection was carried out through direct observation, interviews with the campus, and recording 14 coordinate points of campus service locations using Google Earth. The system is built using the Unity platform with support from the AR Foundation and ARLocation, as well as the A* implementation in route search. The results of this research are able to provide a more efficient, accurate, and interactive campus navigation solution. This system not only supports the convenience of users in finding a location, but also becomes an innovation in the development of campus services based on digital technology.

Application of Forward Chaining and Certainty Factor Methods to Identify Anxiety Disorder Categories

2025-08-21T12:21:23+07:00

Anxiety disorders are a form of mental disorders that often occur and have a significant impact on the quality of life of individuals. However, the process of diagnosing this disorder still faces various challenges, especially limited access to professionals and difficulties in identifying the type of disorder based on varying symptoms. This research aims to design and implement an expert-based system to help the early diagnosis process of anxiety disorders quickly and accurately. The system was developed as a web application that allows users to answer a series of questions related to the symptoms experienced, then provide possible types of disorders based on the calculation of confidence levels. The method used is forward chaining as an inference engine to conduct a rule and certainty factor search to calculate the level of confidence in the identification results of the symptoms experienced by the user. Data collected from the literature and interviews with experts were built into a knowledge base consisting of 8 types of anxiety disorders with a total of 41 symptoms. Each rule in the system is formulated using an if-then structure that combines CF values to represent the level of confidence in the symptoms and the results of logical inference with advanced tracking methods. The system was tested using 20 test data in the form of symptom-based case simulations. The results of the evaluation showed that the system was able to produce an initial diagnosis with an accuracy rate of up to 100% based on comparison with manual diagnosis from experts. This system also provides explanatory information in the form of confidence level in each diagnosis result. These findings suggest that the Certainty Factor and Forward Chaining approaches are effective in building expert systems for diagnosing anxiety disorders and have the potential to be further developed as a screening tool in educational or primary health care settings.

Prediction of Extreme Poverty Levels Using the Performance of the Multiple Linear Regression Method

2025-08-21T14:47:10+07:00

Extreme poverty is a type of poverty that is defined as a condition that cannot meet basic human needs. The Government of Indonesia through Presidential Instruction No. 4 of 2022 sets a target for the elimination of extreme poverty, but this effort requires an accurate and comprehensive data-driven approach. This study aims to build a model for predicting extreme poverty levels. The method used in this study is Multiple Linear Regression (MLR), which is able to measure the contribution of each predictor variable to the phenomenon of extreme poverty. The dataset processed in this study was sourced from the Dumai City Social and Community Empowerment Office. The dataset consisted of 2,007 extreme poverty data with predictor variables in the form of residence ownership (X1), employment (X2), income (X3), education (X4), and health insurance (X5). The results of this study show that the Multiple Linear Regression method is able to provide accurate predictions of the extreme poverty level in Dumai City with an accuracy rate of 87%. The model evaluation was carried out using three metrics based on the results of the test obtained R = 0.674 and R² = 0.454, which means that 45.4% of the variation in poverty status can be explained by the variables of home ownership, type of occupation, amount of income, education level, and health insurance. The ANOVA test showed a value of F = 332.777 with a significance of < 0.001, so the model was simultaneously significant. The regression coefficient showed that all variables had a negative and significant influence (p < 0.05) on poverty status, with the greatest influence coming from the type of job (β = -0.304) and amount of income (β = -0.291), followed by home ownership, health insurance, and education level. Thus, the Multiple Linear Regression method has proven to be effective in building an extreme poverty prediction system. This model can be a basic reference in supporting more targeted, measurable, and data-based socio-economic policy decision-making, especially in efforts to combat extreme poverty in a sustainable and systematic manner.

Application of Fuzzy Logic Method and Analytical Hierarchy Process in Assessment of Education Quality at the Madrasah Aliyah Level

2025-08-17T14:12:43+07:00

Madrasas have a strategic role in producing a generation of the nation that is intellectually intelligent as well as spiritually and emotionally mature, but the quality between madrasas is still uneven, especially in rural areas such as Kerinci Regency. The assessment of the quality of madrasas until now tends to be subjective and has not been based on a measurable, systematic, and widely replicated system. This research aims to develop an objective and structured quality evaluation system for madrasah education using a technology-based approach. The methods used are Fuzzy Logic and Analytical Hierarchy Process (AHP) which are combined into Fuzzy AHP. The scope of the research is focused on aliyah madrasas under the coordination of the Ministry of Religious Affairs of Kerinci Regency. The data was obtained through direct interviews with the Head of the Madrasah Education Section and included eight aliyah madrasas as the object of the research, which were assessed based on nine main criteria: quality of teachers, teaching materials, infrastructure, school governance, learning environment, assessment system, leadership of school principals, parental support, and technology and digitalization. A web-based decision support system was developed to automatically manage Fuzzy AHP calculations, so that it can be used as a continuous evaluation tool. The results of the study show that this model is able to produce consistent, objective, and valid assessments with a Consistency Ratio (CR) value of < 0.1. Madrasah MA2 obtained the highest ranking in the assessment of the quality of education. The Fuzzy AHP approach has proven to be effective in multi-criteria education evaluation and can be the basis for policies that are more responsive to local needs.

Utilization of Convolutional Neural Network Method in Customer Identification Based on Facial Images

2025-08-10T04:02:05+07:00

Artificial intelligence-based facial recognition technology, especially using the Convolutional Neural Network (CNN) method, is increasingly widespread in various business applications, such as customer data management. This technology allows the system to recognize and identify individuals automatically through facial images, so it is very potential to be applied in customer management. This study aims to implement CNN technology in automatically identifying old customers in a case study in JAVApace Studio. CNN method for facial recognition, optimizing the accuracy of old customer identification, designing CNN system integration in computer vision-based applications, and measuring CNN performance in real-time facial identification.

The research method was carried out using a quantitative approach through data collection stages in the form of 875 customer facial images taken in JAVapace Studio, data preprocessing (cropping, resizing, and data augmentation), dataset division for training, validation, and testing. The CNN model used is the ResNet-50 architecture with fine-tuning techniques and freezing layers to improve training efficiency. Model performance evaluation uses a confusion matrix with accuracy, recall, and precision metrics. The results show that the CNN-based facial recognition system achieved 95.7% accuracy in distinguishing existing customers from the test data used. The recall rate was 94.5%, while the precision rate reached 96.2%. The discussion of the results also indicates that the fine-tuning approach is effective in optimizing model performance with an inference time suitable for real-time implementation needs. This study confirms that the implementation of CNN with ResNet-50 architecture is effectively able to recognize the faces of old customers with high levels of accuracy, recall, and precision, making it the right solution in managing customer data automatically and efficiently.

Identification of Skin Diseases in Toddlers Using Convolutional Neural Networks

2025-08-10T04:07:58+07:00

The development of Artificial Intelligence (AI) technology, particularly in the field of computer vision, has made a significant contribution to medical image analysis. Skin disease in toddlers is a common health problem, especially in developing countries. Toddlers' skin is highly susceptible to various infections and dermatological conditions, ranging from bacterial and viral infections to allergies. Some skin diseases frequently found in toddlers include eczema, dermatitis, impetigo, and fungal infections. This study aims to develop a skin disease classification system in toddlers using the Convolutional Neural Network (CNN) method that can be implemented in applications. The Convolutional Neural Network (CNN) method and the U-Net architecture are used to identify skin diseases in toddlers, requiring a fast and accurate diagnosis, but limited medical personnel and examination time are challenges. A deep learning-based system is proposed to assist the automatic identification process. The research dataset consists of 100 toddler skin images obtained from Siti Rahmah Islamic Hospital, covering various types of common skin diseases. The preprocessing process includes cropping, resizing to 128x128 pixels, normalization, and data augmentation to increase the diversity of the dataset. The CNN architecture is used in the feature extraction stage through convolution and pooling layers, while the U-Net is applied in the segmentation stage to separate the wound area from healthy skin with high precision through the encoder-decoder mechanism and skip connection. The model is trained using the Adam optimization algorithm with the Binary Cross-Entropy loss function and the accuracy evaluation metric and Mean Intersection over Union (IoU). The results show that the system is able to segment the wound area with 95.7% accuracy on the test data, and produces fast and efficient detection. The application of the CNN and U-Net methods in this study proves its effectiveness in supporting the medical diagnosis process, especially in cases of toddler skin diseases, as well as can be a reference in contributing to improving the quality of health services, especially in the diagnosis of skin diseases in toddlers and the development of computer vision-based decision support systems in the health sector.