Jurnal Sistemasi (OJS FTIK - UNISI, Fakultas Teknik dan Ilmu Komputer Universitas Islam Indragiri)
Not a member yet
946 research outputs found
Sort by
Adaptive Traffic Signal System Utilizing YOLOv11 and Fuzzy Logic for Congestion Mitigation
Full text linkThe increasing number of vehicles in urban and suburban areas has led to traffic congestion, resulting in longer travel times, higher exhaust emissions, and an increased risk of accidents. Conventional fixed-time traffic signal systems often fail to respond dynamically to changing traffic conditions, leading to inefficient vehicle queues. This study proposes the development of an adaptive traffic signal system that utilizes YOLOv11 and fuzzy logic to detect vehicle volume and adjust green light durations in real time. YOLOv11 is employed to detect vehicles in each lane, while fuzzy logic is used to regulate green signal durations based on the detected vehicle counts. Experimental results demonstrate a detection accuracy of 0.92 and a recall of 0.93. The green light duration varies from 80 seconds for low traffic volumes to 100 seconds for high traffic volumes. The traffic signal cycle is dynamically adjusted according to vehicle density, with a maximum total cycle time of 100 seconds. Overall, the proposed system is proven effective in reducing congestion and improving traffic management efficiency at intersections with high vehicle volumes
A Multimodal Deep Learning Framework for Amyotrophic Lateral Sclerosis Diagnosis using Clinical and Audio Morphology Features
Full text linkAmyotrophic Lateral Sclerosis (ALS) is a highly progressive neurodegenerative disease that impairs motor and speech function. Conventional diagnostic methods, both invasive and non-invasive, are often time-consuming and produce limited sensitivity. This leads to delays in treatment and worsening disease progression. This study proposes a multimodal deep learning framework that utilizes and integrates invasive medical records with non-invasive morphological features of patient speech audio extracted into Mel-Spectrograms. Unlike previous studies that focused solely on speech or clinical features, this study introduces an integrated multimodal diagnostic framework that effectively combines both data sources to achieve reliable diagnostic accuracy. The study included two experimental scenarios. In the first scenario, the audio-trained model used a Convolutional Neural Network (CNN) and was systematically optimized by testing variations in network depth, feature fusion techniques, and layer dropout probabilities to improve model generalization and stability. From the experimental results of the first scenario, the CNN achieved the best performance, achieving 80.33% accuracy in classification using audio data alone from all the tested model variations. In the second experimental scenario, when the best model was trained by incorporating clinical data, the model demonstrated improved diagnostic performance, achieving 100% accuracy. This finding highlights the importance of combining data modalities or sources from various domains, both invasive and non-invasive, to achieve optimal model performance for early ALS detection
Image Forensics Analysis of the Authenticity of Digital Payment Evidence using the K-Nearest Neighbor Algorithm
Full text linkThe rapid growth of digital transactions has also increased the risk of digital payment evidence forgery, such as screenshot manipulation or digital image editing. This study aims to develop an automated authenticity validation system for digital payment evidence by integrating Image Processing, Image Forensics, and Optical Character Recognition (OCR) technologies. The processing pipeline begins with image preprocessing, followed by forensic feature extraction and OCR-based text analysis, which are then classified using the K-Nearest Neighbor (KNN) algorithm.
This study evaluates 15 experimental scenarios based on combinations of training and testing data ratios (90:10, 80:20, 70:30, 60:40, and 50:50) and random state values (42, 32, and 22). Model performance is assessed using accuracy, precision, recall, and F1-score metrics across a range of k values from 1 to 15. The results indicate that the optimal performance is achieved at k = 7, with an accuracy of 97.1%. The proposed system is able to efficiently distinguish between authentic and manipulated digital payment evidence. The system is implemented as an Android application that allows users to upload payment evidence via the device camera or gallery, after which the system automatically analyzes its authenticity. The findings demonstrate that the integration of image forensic techniques and the K-Nearest Neighbor (KNN) algorithm effectively detects indications of manipulation in digital payment evidence and enhances the efficiency of the verification process within the digital financial services ecosystem
Security Mitigation Analysis of Mobile Application Using Static and Dynamic Methods with MobSF
Full text linkThis study evaluates the security of the Mobile Application for the Palm Oil Harvest Information System using static and dynamic analysis through the Mobile Security Framework (MobSF). The research is motivated by the high risk of exploitation in APK-based applications and the lack of in-depth security assessments for applications that manage farmers’ operational data. Static analysis was conducted to identify structural weaknesses, including the use of debug certificates, enabled debugging mode, a low minimum SDK version (minSdkVersion), and exported components without proper protection. The initial results showed an App Security Score of 43/100 (Medium Risk), which increased to 67/100 (Low Risk) after configuration improvements were applied. Dynamic analysis was then performed to assess application security during runtime. The results indicated that the client side was relatively secure, with HTTPS-encrypted communication and no logging of sensitive data. However, dynamic analysis revealed vulnerabilities on the server side, where several backend endpoints could be accessed without authentication and without parameter validation, leading to potential risks of Broken Access Control and Insecure Direct Object Reference (IDOR). The findings confirm that static improvements are effective in strengthening the structural security of the application. Nevertheless, reinforcing authentication, authorization, and request validation mechanisms on the backend API remains essential to ensure comprehensive security before deployment in an operational environment. Unlike previous studies that generally focus only on vulnerability mapping, this study evaluates the effectiveness of security mitigation in a step-by-step manner by demonstrating improvements in static analysis scores and re-validating the results through dynamic analysis. Therefore, this research provides a more comprehensive security assessment of mobile applications by covering both client-side and backend aspects
Evaluation of IT Governance for the Inti Accounting System in Retail SMEs using COBIT 2019
Full text linkThe utilization of information technology (IT) in retail micro, small, and medium enterprises (SMEs) is essential for improving operational efficiency and service quality. However, limited resources and dependence on external vendors pose significant operational risks. Swalayan Pasar Pagi in Tegal utilizes a locally hosted Inti Accounting System supported by external technical services, making it vulnerable to system downtime and weak internal controls. This study aims to evaluate the governance of the system using the COBIT 2019 framework through an analysis of design factors and the governance system design workflow. The research adopts a qualitative descriptive approach, with data collected through semi-structured interviews, observations, and documentation. The results indicate that the supermarket’s strategy focuses on cost efficiency and improving customer service, with primary objectives including enhancing the quality of management information, optimizing business processes, and increasing operational efficiency. Critical risks identified include vendor dependency, reliance on local server infrastructure, and the absence of internal audits. COBIT 2019 mapping identifies BAI10, EDM05, APO14, and APO12 as priority governance and management objectives. The recommended improvements include the establishment of Service Level Agreements (SLAs), regular data backups, periodic system audits, and training for non-IT staff. This study provides practical contributions to strengthening IT governance in retail SMEs and extends the literature on the application of COBIT 2019 in small enterprises, which remain underexplored. The findings also demonstrate that adapting the COBIT 2019 framework is effective in enhancing operational efficiency and reducing IT-related risks in retail SMEs
Mango Leaf Disease Detection using Threshold with CNN ResNet50 Architecture
Full text linkMango leaf diseases pose a significant threat to farmers’ productivity in Indonesia due to the difficulty and inaccuracy of manual diagnosis. A mango leaf disease detection system was developed by optimizing the decision threshold for classification using a ResNet50 Convolutional Neural Network (CNN). The Kaggle dataset consisted of 3,979 mango leaf images across eight classes: healthy, anthracnose, bacterial canker, gall midge, cutting weevil, dieback, sooty mold, and powdery mildew. The raw dataset was processed in Roboflow with an 80:10:10 train-validation-test split, and threefold data augmentation on the training set produced a total of 9,600 images. Decision threshold optimization using the precision-recall curve analysis identified 0.85 as the optimal threshold. At this threshold, precision reached 97.03%, while recall was 94.36%. These results provide a critical reference for agricultural applications in Indonesia, particularly considering local characteristics. The model achieved an F1-score of 95.49% after validation on the augmented dataset specifically tailored for tropical conditions
Improving DCT-based JPEG Steganography using Adaptive LSB Matching for Resistance to Entropy-based Detection
Full text linkData transmission security has become a critical issue in the digital era, where steganography plays an important role in concealing confidential information within digital media. A key limitation of conventional Discrete Cosine Transform (DCT)-based steganography in JPEG images is its vulnerability to statistical detection through entropy analysis, as well as the risk of significant degradation in visual quality. This study aims to enhance DCT-based steganography techniques to minimize entropy-based detection while maintaining an optimal Peak Signal-to-Noise Ratio (PSNR). The proposed method employs an Adaptive LSB Matching approach by embedding messages into low-to-mid frequency coefficients using an adjustment mechanism (x±1). The performance of this method is then compared with the standard DCT approach. Experimental results show that the proposed method is able to preserve visual quality, achieving an average PSNR of 40.41 dB under maximum payload conditions, while reducing the entropy difference (ΔH) to 0.00251. These findings demonstrate that the developed technique is more robust against statistical steganalysis attacks and provides better visual fidelity compared to conventional methods
Hyperparameter Optimization of Ensemble Learning for Heart Disease Prediction using Patient Data
Full text linkThis study evaluates the impact of hyperparameter optimization on the performance of four machine learning algorithms—Extra Trees, XGBoost, Random Forest, and AdaBoost—in heart disease prediction. The results show that hyperparameter tuning significantly improves model performance for three out of the four algorithms, with varying effects across models. Extra Trees demonstrates the most consistent improvement, achieving the highest Area Under the Curve (AUC) of 0.9107 and a recall of 80.93%, which is particularly crucial in medical contexts for accurately identifying disease cases. XGBoost exhibits the largest increase in accuracy, rising from 78.11% to 81.49%, while Random Forest shows improvements in both recall and F1-score. In contrast, AdaBoost experiences a slight decline in performance, suggesting that the model was already near optimal prior to tuning. Overall, Extra Trees with hyperparameter optimization emerges as the best-performing algorithm for heart disease prediction, offering high reliability in identifying at-risk patients
Implementation of Random Forest for Predicting Complaint Handling Priorities at the Minahasa Education Office
Full text linkPublic service delivery in the education sector requires government institutions to handle public complaints promptly and accurately. However, the Minahasa Regency Education Office still faces challenges in determining complaint handling priorities, which are often subjective and manually assigned. This condition may lead to delays in addressing urgent complaints. This study aims to implement the Random Forest algorithm to objectively predict complaint handling priorities based on data. The research methodology includes collecting a dataset of 500 public complaints with six attributes: complaint type, violation level, work unit, follow-up action, outcome, and priority. The process involves data preprocessing, splitting the dataset into training and testing sets, building the Random Forest model, and evaluating its performance. Data processing and modeling were conducted using Jupyter Notebook within the Anaconda environment. Model performance was evaluated using accuracy, confusion matrix, precision, recall, and F1-score metrics. The results show that the Random Forest algorithm achieved an accuracy of 100%, with precision, recall, and F1-score values of 1.00 across all priority classes. These findings indicate that the model demonstrates excellent and stable classification performance. Therefore, it can serve as a foundation for developing a decision support system to improve the effectiveness and quality of public service delivery at the Minahasa Regency Education Office
Analysis of User Experience in the WEVERSE Application using the ITIL V3 Service Design Domain : An ENGENE Perspective in Indonesia
Full text linkThe rapid development of digital technology has changed the way artists and fans interact through community applications such as Weverse. This study aims to evaluate the user experience of Weverse from the perspective of ENGENE (ENHYPEN fans) in Indonesia using the ITIL V3 Domain Service Design approach, as well as to analyze the service quality of the application and provide recommendations for improvement. The research employed a quantitative approach through a survey using a questionnaire instrument developed based on the ITIL V3 Service Design Framework, which consists of six subdomains : Service Catalogue Management, Service Level Management, Capacity Management, Availability Management, Information Security Management,and Supplier Management. The sample consisted of 100 respondents selected using the Slovin formula with a 10% margin of error from an estimated population of 29,937 ENGENE members in Indonesia. Data were analyzed using maturity level analysis and gap analysis. The results show that all subdomains are at Maturity Level 4 (Managed and Measurable), with an average current maturity score of 4.07 and Expected Maturity score of 4.50, resulting in an average gap of 0.43. Among the subdomains, Service Catalogue Management demonstrates the best performance with a current maturity score of 4.21 and the smallest gap of 0.29, while Service Level Management shows the lowest score of 3.89 with the largest gap of 0.61, making it the primary priority for improvement. The findings suggest that the Weverse application has reached the managed stage, where IT service design processes can be systematically measured and monitored. However, continuous improvements are still required to achieve the optimal target of Maturity Level 5 (Optimized)