JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer)

PARKINSONS DISEASE DETECTION USING INCEPTION AND X-CEPTION WITH ATTENTION MECHANISM

Eka Rahma Agustina — 2025-08-15

Parkinson's disease is one of the global health challenges that requires early detection to slow the progression of symptoms. This study proposes an automatic detection system based on deep learning using the InceptionV3 and Xception architectures combined with a multi-head awareness mechanism. The dataset used consists of 72 handwritten spiral images, comprehensively distributed between the Healthy and Parkinson's categories. The process includes preprocessing in the form of normalization and image resizing, as well as model training using the Adam algorithm and the binary cross-entropy loss function. The results show that the model is able to classify both categories with high accuracy. The use of the attention mechanism provides a performance increase of 4.2% on InceptionV3 and 3.1% on Xception compared to the version without attention. In data testing, the InceptionV3 model with attention achieved 100% accuracy, 100% precision, 100% recall, and 100% F1-score. Meanwhile, the Xception model with attention achieved 88% accuracy, 90% precision, 88% recall, and 87% F1-score. The attention mechanism also helps the model in capturing important features such as vibration and irregularity of the spiral pattern. This research makes an important contribution to the development of an artificial intelligence-based automated early diagnosis system to detect Parkinson's disease more accurately and responsively.

PREDICTION MODEL OF HUMAN DEVELOPMENT INDEX (HDI) USING K-NEAREST NEIGHBOR (KNN) ENSEMBLE

Fitri Nuraeni — 2025-08-15

The Human Development Index (HDI) is an essential indicator in measuring the success of human development. Although some regions in Indonesia have experienced increased HDI, inequality between areas makes it difficult to predict future HDI values. This research aims to build an HDI prediction model using the ensemble K-nearest neighbor (KNN) method. The dataset consists of 574 data points with attributes of life expectancy, expected years of schooling, average years of education, and regional income per capita. The method used is SEMMA with z-score normalization, feature selection based on domain knowledge, and validation with 10-fold cross-validation. The results showed that the KNN Ensemble model with the Boosting (Adaboost) technique had the best performance with an average MAPE of 0.58%, which indicates that the model's predictions deviate by less than 1% from actual HDI values, which is considered highly accurate and reliable for policy planning. This model proved better than linear regression, neural networks, single KNN, and double exponential smoothing algorithms. The improved prediction accuracy of the proposed model provides local governments with a reliable tool for scenario-based development planning and policy simulation, contributing to achieving the Golden Indonesia 2045 strategic vision.

GFPGAN UPSCALING FOR HUMAN FACIAL EXPRESSION CLASSIFICATION USING VGG19 ARCHITECTURE

Andhika Rezky Fadillah — 2025-08-15

Human facial expression recognition is a rapidly evolving field in artificial intelligence and digital image processing. This study aims to develop a model capable of recognizing and classifying human emotions through facial feature analysis. However, a major challenge in facial expression classification is low image quality, which can reduce model accuracy. Factors such as poor lighting, low resolution, variations in viewing angles, and occlusion (obstructions) on the face pose significant obstacles to accurate detection.This research proposes the application of an upscaling method using the Generative Facial Prior Generative Adversarial Network (GFPGAN) to enhance facial image quality by restoring details in expressions that may be unclear due to low resolution. After the upscaling process, facial expression classification is conducted using a CNN architecture based on VGG19, and the model is evaluated using accuracy, precision, recall, and F1-score metrics to assess its performance in emotion detection. Experiments are conducted in two scenarios: classification without upscaling and classification with GFPGAN upscaling. The results indicate that integrating GFPGAN with the VGG19-based CNN proposed in this study significantly improves emotion detection accuracy, achieving 86%, compared to 76% for the model without image quality enhancement

THE IMPACT OF COLOR AND CONTRAST ENHANCEMENT FOR DIAGNOSING GASTROINTESTINAL DISEASES BASED DEEP LEARNING

Gregorius Guntur Sunardi Putra — 2025-08-15

Endoscopy is a crucial tool for diagnosing digestive tract diseases—colon cancer and polyps using a camera with LED lighting, but often results in low-quality images with poor contrast and luminance. This study evaluates the performance of two contrast-based image quality enhancement—Contrast Limited Adaptive Histogram Equalization (CLAHE) and Improved Adaptive Gamma Correction with Weighting Distribution (IAGCWD)—along with various color space transformations (RGB, HSV, YCbCr, CIELAB, Grayscale) in deep learning-based digestive tract diseases detection system. The detection system using EfficientNetV2S model and Quadratic Weighted Kappa (QWK) loss function to obtain the balance of prediction results for each class. The experiment shows that CLAHE is able to achieve 79% accuracy which is superior in clarifying important information in endoscopy images. CLAHE performs well due to its ability to reduce noise and enhance contrast. The classification model with HSV and CLAHE on KVASIR is able to recognize all classes well. RGB, HSV, and YCbCr color spaces have stable performance in most tests. This study contributes insights for enhancing endoscopic image quality to support both computer-aided and clinical diagnosis.

PRESERVATION OF THROUGH PATTERN RECOGNITION USING A COMBINATION OF GLCM, LBP, AND SVM MULTICLASS

Budiman Baso — 2025-08-15

This study proposes an automatic method to recognize traditional Timorese weaving patterns using machine learning techniques. Timorese weaving image data is processed through pre-processing stages and its features are extracted using the Gray Level Cooccurrence Matrix (GLCM) and Local Binary Pattern (LBP) methods, which function to capture the characteristics of texture and design in the weaving patterns. The classification model is built with the Support Vector Machine (SVM) algorithm using the One Versus One (OVO) and One Versus All (OVA) approaches with several kernels, including Linear, Polynomial, and Radial Basis Function (RBF). The best results were obtained with the Linear kernel and the OVO method, resulting in an accuracy of 88.66%, a precision of 88.66%, a recall of 88.80%, and an F1-score of 88.73%. This approach shows great potential in preserving and documenting Timorese weaving patterns automatically and efficiently, with accurate classification results. This study explores a machine learning approach for identifying Timorese weaving patterns. By leveraging GLCM and LBP for texture analysis and SVM with OVO and OVA for classification, the method achieves high accuracy. The findings support digital preservation efforts and cultural heritage conservation.

OPTIMIZING IT GOVERNANCE FOR ENHANCED SECURITY IN SMART CITIES

Agustinus Fritz Wijaya — 2025-08-15

The rapid digitization of urban environments through technologies such as the Internet of Things (IoT), cloud computing, and big data analytics has significantly transformed modern cities into smart cities. However, this transformation has raised critical concerns regarding the security and privacy of citizen data. Prior studies have explored various IT governance models, yet there remains a gap in their contextual application to the dynamic and complex nature of smart cities. This research addresses that gap by examining the strategic role of Information Technology (IT) governance in enhancing data security and privacy in smart city initiatives. Through a literature review and analysis of case studies, this study identifies key IT governance frameworks and best practices, and adapts them to the unique operational, regulatory, and infrastructural demands of smart cities. The findings reveal that aligning IT governance with institutional policies, risk management, and legal compliance significantly strengthens urban digital resilience. Moreover, the incorporation of real-time monitoring systems, encryption protocols, and structured incident response plans is found to be effective in mitigating cyber threats. The novelty of this study lies in its integrated model that combines governance principles with smart city-specific risk contexts, offering a strategic roadmap for policymakers. This research contributes to the development of adaptive governance strategies that not only ensure compliance and security but also build public trust in digital urban services. Limitations of the study include the reliance on secondary data and the need for empirical validation, which will be addressed in future research through pilot implementations and stakeholder engagement.

ENHANCING MACHINE LEARNING ALGORITHM PERFORMANCE FOR PCOS DIAGNOSIS USING SMOTENC ON IMBALANCED DATA

Rofiqoh Dewi — 2025-08-15

Polycystic Ovarian Syndrome (PCOS) is one of the most frequently occurring endocrine disorders in women of reproductive age, distinguished by disruptions in hormonal regulation that can impact menstrual cycles, fertility, and physical appearance. Despite its high prevalence, PCOS is often diagnosed late and inaccurately, leading to inappropriate treatment and long-term health issues for patients. Machine learning can serve as an effective solution to enhance the accuracy of PCOS diagnosis. However, one of the primary challenges encountered is the class imbalance in the dataset, where the number of positive case data (PCOS) is often significantly lower than the negative case data. This imbalance can result in a biased model that is less effective in predicting the actual condition of patients. In this study, the Synthetic Minority Over-sampling Technique for Nominal and Continuous (SMOTENC) method is recommended to address the issue of imbalanced data, thereby improving the performance and accuracy of the machine learning model employed. The evaluation matrix test results clearly demonstrate that the accuracy of each machine learning model improved after applying the SMOTENC method. Specifically, the accuracy of the K-Nearest Neighbors (KNN) algorithm increased from 81.6% to 89.8%, the Support Vector Machine (SVM) algorithm from 90.6% to 92.5%, the Naive Bayes algorithm from 70% to 82.3%, and the C4.5 algorithm from 99.6% to 99.7%. This research provides a substantial contribution to advancing the development of diagnostic methods thatare both more precise and efficient.

FINE-GRAINED SENTIMENT ANALYSIS ON BIG DATA FROM MULTI-PLATFORM IN INDONESIA

Ronsen Purba — 2025-08-15

Sentiment analysis on multi-platform big data in Indonesia presents a complex challenge, particularly in optimizing sentiment classification with higher granularity levels. This study aims to develop and optimize a sentiment classification model for analyzing public opinion on ChatGPT using a Fine-Grained Sentiment Analysis approach based on Indonesian Bidirectional Encoder Representations from Transformers (IndoBERT). The method is applied to big data collected from various social media platforms to improve accuracy and precision in identifying a broader spectrum of sentiments, including highly positive, positive, neutral, negative, and highly negative categories. A comparative analysis was conducted on different base models, including BERT, RoBERTa, and IndoBERT, to determine the most effective model. Experimental results show that the optimized IndoBERT model achieves an accuracy of 96% and outperforms other models in terms of precision and F1-score across all sentiment categories. Additionally, this study evaluates the model's computational efficiency and adaptability to diverse data. Thus, the developed model can serve as a more effective solution for gaining deeper insights into public opinion across various digital platforms in Indonesia.

ENHANCING SENTIMENT ANALYSIS ACCURACY WITH BERT AND SILHOUETTE METHOD OPTIMIZATION

Kelvin Kelvin — 2025-08-15

This research is based on the emergence of ChatGPT technology, which has significant implications in various fields. This research aims to design a model that improves sentiment analysis classification accuracy. The methods applied include the use of the Silhouette Coefficient to determine the best cluster parameters before performing data grouping with the Self-Organizing Map (SOM) method. Additionally, the Bidirectional Encoder Representations from Transformers (BERT) model is utilized to perform precise and convergent sentiment classification. The research methodology encompasses several phases, including data preprocessing through natural language processing techniques. Textual data is converted into vector representations, which are then processed using the Silhouette Coefficient to identify the optimal cluster parameters. These parameters are subsequently applied in the Self-Organizing Map method to cluster data, while the Bidirectional Encoder Representations from Transformers model determines public sentiment, categorized as positive, negative, or neutral. The findings of this study indicate that the best cluster parameter is 9, using a batch size of 64 and a maximum sequence length of 128. The highest accuracy achieved using the confusion matrix is 92.06%. Further tests with varying parameters confirm that the Silhouette Coefficient method significantly enhances the convergence and accuracy of classification outcomes. The conclusion of this research is that integrating the Silhouette Coefficient and Bidirectional Encoder Representations from Transformers is effective in optimizing sentiment analysis on large datasets, achieving both accurate and reliable results.

PERFORMANCE COMPARISON OF DEEP CNN ARCHITECTURES FOR LUNG AREA SEGMENTATION IN CHEST IMAGING

Larissa Navia Rani — 2025-08-15

Lung area segmentation is a critical preprocessing step in computer-aided diagnosis systems for respiratory diseases such as lung cancer and pneumonia. Accurate segmentation enhances the detection and monitoring of pathological conditions but manual delineation is time-consuming and subject to variability. This research aims to identify the most effective convolutional neural network (CNN) architecture for automated lung segmentation by comparing three models: U-Net, DeepLab, and a proposed hybrid model combining U-Net with ResUNet_Light. The models were trained and evaluated using a publicly available chest CT dataset under identical experimental settings, including preprocessing steps, training parameters, and standard evaluation metrics: Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Precision, and Recall. Results show that the proposed U-Net + ResUNet_Light model achieves the best performance across all metrics (DSC: 0.6767, IoU: 0.5652, Precision: 0.8480, Recall: 0.7920), outperforming both U-Net and DeepLab. These improvements are attributed to the integration of residual blocks, which enhance feature propagation and gradient flow, enabling better generalization and segmentation accuracy, especially along complex lung boundaries. In contrast, while DeepLab performs well in capturing contextual information, its higher complexity may hinder real-time applicability. U-Net, though efficient, showed limitations in accurately segmenting irregular regions. The findings demonstrate the potential of the proposed model for clinical deployment, where both accuracy and efficiency are critical. This study contributes to the development of more robust deep learning-based segmentation methods and highlights the importance of architectural enhancements in CNN design for medical image analysis.

COMPARATIVE PERFORMANCE STUDY OF SEARCH ALGORITHMS ON LARGE-SCALE DATA STRUCTURES

purnama182 Nyoman Purnama — 2025-08-15

— In the era of big data, searching for information in big data sets is a big challenge that requires efficient search algorithms. This study compares the performance of three classic search algorithms, namely linear search, binary search, and hash search. This study uses large-scale datasets, namely Amazon Product Reviews and Amazon Customer Reviews. Evaluations were conducted based on the complexity of time for each search method. The results of the experiment showed that linear search had the slowest performance with O(n) time complexity, making it inefficient for large data sets. Binary search performs better with O(log n) complexity, but requires pre-sorted data. Hash searches provide the most optimal results in best-case and average with O(1) complexity, but can be reduced to O(n) in the worst case when there are too many collisions in the hash function. Hash search consistently outperforms linear and binary searches in terms of execution speed. Binary search remains highly efficient for sorted data, while linear search is clearly the least efficient, especially for large-scale datasets. Linear search has high execution times and is inconsistent, while binary and hash search are more efficient and stable. The algorithm's performance did not differ significantly between datasets, suggesting the data structure did not affect performance as long as the search type was the same.

PARAMETER TUNING IN BACKPROPAGATION NEURAL NETWORKS: IMPACT OF LEARNING RATE AND MOMENTUM ON PERFORMANCE

Syaharuddin Syaharuddin — 2025-08-22

Artificial Neural Network (ANN) play a pivotal role across diverse domains, including medicine, economics, and technology, due to their ability to model complex relationships and deliver high prediction accuracy. This study systematically examines how learning rate and momentum interact in backpropagation, moving beyond isolated analysis to enhance ANN performance. A qualitative research design employing a systematic literature review was utilized, with data sourced from reputable databases covering the past 11 years. Bibliometric tools such as VOSviewer and R-Studio were applied to identify trends and patterns in the literature. The findings reveal that both learning rate and momentum significantly impact convergence efficiency and model stability. Backpropagation remains fundamental for weight adjustment in minimizing prediction errors. ANN optimization demonstrates substantial practical benefits, including enhanced treatment outcome predictions in medicine, modeling nonlinear patterns in economics, and improved image classification accuracy. However, challenges such as the curse of dimensionality, overfitting, and dependence on large datasets persist. Strategies such as regularization, ensemble methods, and sensitivity analysis present viable solutions. This study underscores the critical need to advance ANN optimization techniques and highlights the potential of interdisciplinary collaboration in addressing existing limitations and broadening ANN applications

COMPARATIVE ANALYSIS OF CNN ARCHITECTURES FOR TOMATO LEAF DISEASE CLASSIFICATION USING TRANSFER LEARNING

Anton Anton — 2025-08-22

Tomato is one of the widely available horticultural products and holds significant economic value in Indonesia. However, its productivity is often disrupted by various leaf diseases. This study aims to compare the performance of three CNN architectures—DenseNet121, Xception, and MobileNetV2—in classifying tomato leaf diseases. The dataset used consists of 10,000 balanced images across ten classes: Bacterial Spot, Septoria Leaf Spot, Early Blight, Late Blight, Mosaic Virus, Yellow Leaf Curl Virus, Leaf Mold, Target Spot, Spider Mites Two-Spotted Spider Mite, and Healthy. All images were resized to 224x224 pixels and divided into 80% training data and 20% test data. Augmentation techniques were applied to balance the data across classes. Experimental results show that the Xception architecture outperforms the other models, achieving an accuracy of 98.79%, with a precision of 98.80%, recall of 98.79%, and an F1-Score of 98.78%. These findings indicate that the Xception model is highly effective for plant disease classification and is suitable for implementation in environments with limited resources.

HYBRID LEARNING STRATEGY COMBINING MODEL-LEVEL TRANSFER LEARNING AND DATA-LEVEL AUGMENTATION FOR BRAIN CANCER CLASSIFICATION

Budiman Budiman — 2025-08-22

Due to the complexity of images, size, and balance of data, brain cancer diagnosis is still one of the most challenging problems to solve. It is shown that traditional classification methods based on 'first principles' do not produce ideal results, often due to different brain tumours. This research uses a hybrid model that leverages transfer learning with data augmentation and AI refinement to categorise three brain tumours: glioma, meningioma, and others. This research aims to improve the classification performance of brain cancer detection using this model. The methodology uses a framework created with a specific dataset, mixed data enhancement, and InceptionV3 model refinement to improve performance. With a validation accuracy of 0.95, the F1 scores for glioma, meningioma, and other brain tumours were 0.98, 0.95, and 0.92, respectively. This hybrid model achieves accuracy without complexity in design while addressing data scarcity and balance issues. The primary focus of this research was to create an effective and robust model for classifying brain cancers that is easy to use in low-resource clinical environments. The results demonstrate how deep learning can improve diagnostic precision and provide a scalable method for detecting brain cancer in the early stages of medical imaging

FACIAL RECOGNITION SYSTEM FOR DISTANCE LEARNING STUDENT ATTENDANCE MANAGEMENT USING MACHINE LEARNING

Agus Sriyanto — 2025-08-26

The administration of student attendance constitutes a vital component of academic governance, affecting both educational outcomes and institutional efficacy. Utilizing machine learning to augment precision and efficacy, with adaptability for both physical and remote learning environments. The research methodology encompasses the acquisition of facial data from students under diverse lighting conditions, perspectives, and remote settings, succeeded by preprocessing and training of a facial recognition algorithm employing machine learning techniques. The system addresses key technical challenges such as camera quality variations, lighting inconsistencies, and spoofing risks by integrating robust image preprocessing and security safeguards. Quantitative evaluation shows that under ideal and controlled conditions, the system achieves up to 100% accuracy with an average processing time of 0.8 seconds. With the specifications Intel Core i5, RAM8 GB, minimum windows 10, NVIDIA GeForce GTX 1050, 1080p minimum camera with 30 fps frame rate, standard CMOS sensor, and automatic exposure adjustment capabilities, accuracy will drop if the conditions are not ideal. The system ensures the security and privacy of student facial because it is live with zoom or LMS. Furthermore, the incorporation of this system facilitates the realization of smart campus initiatives by delivering precise, real-time attendance information. This inquiry contributes to educational technology, enhancing operational efficacy and fostering digital transformation within higher education institutions. The designed system also seeks to reduce overall student attendance fraud.

EVALUATING REGRESSION AND NEURAL NETWORKS FOR FIVE TRAIT TEXT-BASED PERSONALITY PREDICTION

Anggit Dwi Hartanto — 2025-08-26

The aim of this study is to evaluate the effectiveness of several predictive modeling techniques in mapping the five major personality traits (extraversion, neuroticism, agreeableness, conscientiousness, and openness) from text-based data. The dataset consists of text-based features extracted from publicly available social media posts, providing a realistic basis for personality prediction. Performance was measured using mean absolute error (MAE), mean squared error (MSE), and R² score to evaluate prediction accuracy and generalization quality, along with training time for computational efficiency. The research compares linear regression, ridge regression, random forest, and neural networks implemented in PyTorch. Results indicate that ridge regression and random forest outperform linear regression and neural networks in error metrics and explained variance, with ridge regression offering a favorable balance between accuracy and training time. Random forest achieves slightly better accuracy but with significantly longer training duration, reducing its practicality for real-time use. Despite theoretical advantages in modeling non-linear relationships, neural networks showed suboptimal results, likely due to limited hyperparameter tuning and dataset size. These findings highlight trade-offs among model complexity, accuracy, and efficiency, suggesting ridge regression as a pragmatic choice for current personality prediction from text while encouraging future research on advanced neural architectures and enhanced datasets

OPTIMIZATION OF STUNTING INFANT DATA CLUSTERING WITH K-MEANS++ ALGORITHM USING DBI EVALUATION

Efmi Maiyana — 2025-08-27

Stunting in infants is a serious health issue, particularly in developing countries like Indonesia. This study aims to optimize the clustering of stunting data in infants using the K-Means++ algorithm, evaluated with the Davies-Bouldin Index (DBI) to determine the optimal number of clusters. The stunting data includes variables such as age, gender, weight, and height. The analysis results indicate that the optimal number of clusters is 5, with a DBI value of 0.837986204, confirming the quality of the clustering. This conclusion demonstrates that the combination of these evaluation methods produces effective clustering and provides significant insights into identifying groups of infants with varying stunting risk levels. These findings can serve as a basis for more targeted health interventions in addressing stunting

DRIP INFUSION MONITORING AND DATA LOGGING SYSTEM BASED ON YOLOv5

Giri Wahyu Wiriasto — 2025-08-27

Intravenous infusion (IV) functions to deliver medication or fluids directly into the patient’s body and requires an accurate drops-per-minute (TPM) calculation to ensure the correct dosage is administered. Manual calculation techniques, which are still widely used today, tend to be inefficient and carry a high risk of human error. Therefore, a more reliable and innovative automated approach is needed. In this study, we developed a prototype of an automatic infusion monitoring system based on the CNN-YOLOv5 architecture. The system records a one-minute IV drip video using a mobile device, then processes it through a server to automatically calculate the TPM, where YOLOv5 is used for drip detection, Deep SORT for object tracking, and a unique ID numbering scheme is applied to each droplet to ensure it is counted only once until it exits the frame. The calculation results are stored in a patient database that we designed. We also explored the effect of dataset background on accuracy. Testing was conducted on 48 videos (30 fps) with two background types—white (LBP) and black (LBH)—and drip variations of 20, 30, 40, and 50 TPM with varying durations. The results showed higher accuracy on the black background, reaching 0.79 compared to 0.58 on the white background, both with a precision of 1.00. The system demonstrated excellent performance in detecting drips with high precision and good accuracy, particularly on LBP for TPM <40 fps and on LBH for TPM <50 fps.

ENHANCED FLOWER IMAGE CLASSIFICATION USING MOBILENETV2 WITH OPTIMIZED PERFORMANCE

Ade Ismiaty Ramadhona Ht Barat — 2025-08-27

Flower classification is an essential activity in multiple fields, including healthcare, cosmetics, agriculture, and environmental monitoring. Deep learning has achieved notable success in intricate picture categorization problems, especially through the utilization of lightweight convolutional neural network (CNN) architectures like MobileNet and MobileNetV2. This work assesses and contrasts the efficacy of four prevalent optimizers Adam, RMSProp, SGD, and Nadam on datasets of flower and herbal leaf images. Experiments were performed using a uniform training configuration on a CPU-based system devoid of GPU acceleration, evaluating both model efficacy and computational efficiency. Evaluation criteria including accuracy, precision, recall, F1-score, and loss were utilised, augmented by confusion matrix analysis. The findings indicate that MobileNetV2 regularly surpasses the baseline MobileNet, with RMSProp attaining the highest accuracy (99.52%) and the lowest loss (0.0126) on the herbal dataset. In the flower dataset, RMSProp achieved the highest accuracy of 96.67%. Moreover, MobileNetV2 necessitated increased memory and extended training duration, while delivering superior classification performance overall. These findings underscore the significance of optimizer selection and model architecture in lightweight deep learning applications, especially for deployment on resource-limited devices.

DEVELOPMENT OF INTEGRATED DIGITAL HR SYSTEM USING WATERFALL FOR LEAVE AND REPORT MANAGEMENT

Firman Maulana Syabani — 2025-08-27

Digitalization in human resource management (HRM) is crucial for modern companies. Several previous studies have discussed HRM, but only a few have focused explicitly on developing leave applications and daily reports, especially regarding comprehensive data recapitulation. Until now, no research has explicitly designed an integrated HRM model incorporating these essential features. In response to this gap, this study presents the design and development of a digital HR management system named “CUTI DULU” to manage leave requests and daily reports. The system was developed Using the Waterfall SDLC model on the Laravel framework to streamline development. It incorporates role-based access control (RBAC) for user permissions and employs JSON Web Tokens (JWT) for secure authentication; data transmission and storage are protected by industry-standard encryption protocols. User Acceptance Testing (UAT) by IT experts and employees verified that the system met functional requirements. Results showed that average leave request processing time fell from four days to about five hours, and administrative errors (e.g. duplicate entries or miscalculations) decreased by 80.95%. Survey responses indicated high satisfaction: 94% of IT experts and 92% of end users reported that the system met their needs. By automating leave and reporting workflows, the proposed system significantly improves administrative efficiency, data transparency, and HR process accuracy. While these results are promising, the current study is limited to a single organizational context, and its broader applicability remains to be validated. Future work should investigate its adaptability across diverse institutional settings to confirm its scalability and generalizability.

SAP ASSESSMENT USING COBIT 2019 AND ITIL FOR SYSTEM IMPROVEMENT AND STRATEGIC DECISION SUPPORT

Joe Yuan Mambu — 2025-08-27

The increasing reliance on Information Technology (IT) for enhancing business performance has led organizations to adopt structured governance and service management frameworks. This study evaluates the IT governance implementation at PT. Natural Indococonut Organik—an organic coconut enterprise that relies on SAP as its core enterprise system. Using the COBIT 2019 and ITIL V.3 frameworks, the study assesses IT process capability, service management maturity, and alignment with best practices. A qualitative descriptive approach was applied through three structured interviews with IT personnel. The first interview used COBIT 2019 Design Factors to identify priority processes: APO12 (Managed Risk), DSS01 (Managed Operations), and MEA03 (Managed Compliance). The second assessed these processes’ capability levels, revealing gaps below the target level (Level 4): APO12 at 33%, DSS01 at 75%, and MEA03 at 12.5%. The third interview applied the ITIL Self-Assessment to evaluate the service desk, with results indicating partial achievement and an overall maturity near Level 2. Key deficiencies were found in risk management, compliance oversight, operational consistency, and user feedback mechanisms—areas critical to supporting SAP effectively. Findings are categorized into design, evaluation, and improvement domains, demonstrating how governance analysis contributes to enhancing enterprise information systems. Strengthening SAP-related risk controls, service procedures, and user engagement processes is essential to elevate governance maturity and system performance