Jurnal Riset Informatika

CLASSIFICATION OF COFFEE LEAF SPOT DISEASES USING THE RESIDUAL NEURAL NETWORKS

2025-10-21T07:42:12+00:00

Coffee is one of the competitive commodities that requires detailed quality control. The common diseases that attack coffee plants are miner, rust, and phoma. Despite their visual similarity, the diseases differ in symptoms and treatments, requiring precise identification aided by computer vision. Miner and phoma have similar image features that are challenging in this study. Avoiding treatment error, several deep learning approach is needed to help classify the diseases. One of the robust methods is the Residual Network. Considering the number of datasets and alignment with the state-of-the-art, this study picked ResNet50 and ResNet101 to be observed. This study employed ResNet50 and ResNet101 in two scenarios. The first scenario was training the models on datasets without preprocessing, while the second scenario trained models on processed datasets. The preprocessing involved converting the color model to HSV and taking the range of leaf spot color from light red to dark brown for color segmentation. This study successfully achieved accuracy, precision, and F1-score at 89,16%, 89,42%, and 89,15% respectively, for the ResNet50 model trained on preprocessed data, slightly higher than the metrics of ResNet101. The ResNet101 achieved 87.95% of accuracy, 88.05% of precision, and 87.98% of F1-Score. These results indicate that ResNet50 is more robust for classifying the leaf spot, and the color segmentation helped the model to optimize the performance

OBJECT DETECTION FOR LOW-LIGHT ENVIRONMENT USING MULTISCALE RETINEX

2026-02-09T21:17:45+00:00

Object detection is a critical task in computer vision, yet its performance degrades significantly under low-light conditions due to loss of detail and diminished features. This study proposes an image enhancement framework to improve detection robustness in challenging lighting. The methodology integrates Multiscale Retinex (MSR) for image enhancement and SSD MobileNet V2 for object detection. MSR was configured with optimal parameters (scale1:10, scale2:60, scale3:180, σ:100, β:30) to enhance brightness while preserving crucial image details. The experimental results demonstrate that Retinex correction is highly effective in extreme low-light scenarios. In 0 lux conditions, where objects were completely undetectable without processing, the proposed method enabled detection with confidence levels between 62% and 96%, yielding an average accuracy increase of 50%. In 15 lux conditions, accuracy improved by 6.6%. However, the system degraded at intensities above 25 lux, suggesting that the enhancement is most beneficial in near-dark environments. In conclusion, Multiscale Retinex significantly enhances the capability of SSD MobileNet V2 for object detection in environments with illumination below 77 lux. This approach provides a viable solution for improving the reliability of surveillance and automated systems operating in unpredictable lighting.

A SEASONAL IMPUTATION METHOD FOR ADDRESSING MISSING DATA IN ENVIRONMENTAL IOT SENSOR TIME SERIES

2026-03-04T03:21:02+00:00

Missing and incomplete observations in Environmental IoT sensor networks reduce data reliability and disrupt analyses, especially for temperature and humidity time series exhibiting strong diurnal seasonality. This study develops and evaluates a seasonal imputation method to address missing data in IoT-based environmental monitoring, using a workflow of anomaly detection, outlier removal, time-of-day-aware imputation, and performance evaluation under varying missing-rate scenarios. Key challenges include sensor noise, connectivity issues, and intermittent hardware failures, which degrade data integrity and affect trend analysis, forecasting, and anomaly detection. To mitigate these, the method uses hourly and minute-level seasonal patterns after filtering out physically unrealistic values. Experimental results show high accuracy and robustness in reconstructing temperature and humidity data: temperature imputation achieves MAE values of approximately 0.86–0.87°C, and humidity yields MAE values of 3.92–4.01%RH, with no performance drop even at 50% data loss. The imputed series preserves natural diurnal dynamics without introducing distortions, effectively restoring continuity and structural consistency in environmental IoT time series for reliable modeling, feature extraction, and decision support.

TOPIC MODELING OF PUBLIC DISCOURSE ON TWITTER ABOUT THE ASSET CONFISCATION BILL USING LATENT DIRICHLET ALLOCATION (LDA)

2026-02-10T02:53:14+00:00

This study examines the structure of public discourse on Twitter regarding the Indonesian Asset Confiscation Bill, a policy initiative aimed at strengthening anti corruption enforcement and ensuring legal certainty. Moving beyond conventional sentiment classification, this research identifies how substantive public concerns are thematically organized within digital debate. A total of 14,319 cleaned and deduplicated tweets collected between January and September 2025 were analyzed using Latent Dirichlet Allocation with the optimal model configuration of nine topics selected based on coherence evaluation to ensure semantic interpretability. The findings reveal nine dominant thematic clusters, with law enforcement and regulatory enactment emerging as the primary focus, followed by legislative process dynamics, protest mobilization, party politics, and institutional accountability. These results indicate that online discourse is structured around normative concerns, particularly procedural clarity, fairness, and institutional legitimacy, rather than driven solely by emotional polarity. Scientifically, this study contributes by shifting the analytical emphasis from sentiment polarity toward systematic thematic mapping of digital political discourse using an optimized LDA framework tailored to Indonesian Twitter data characteristics. Practically, the findings provide policymakers with an evidence based monitoring instrument to identify priority public concerns, strengthen legislative communication strategies, and reduce interpretive ambiguity in sensitive regulatory deliberations.

SENTIMENT CLASSIFICATION OF PUBLIC PERCEPTIONS ON RP200 TRILLION HIMBARA STIMULUS USING NAÏVE BAYES

2026-02-25T01:27:02+00:00

The government's policy in the form of a fund stimulus of Rp200 trillion to the Himpunan Bank Milik Negara (HIMBARA) is a strategic step to maintain national economic stability and encourage real sector recovery. However, the implementation of public policy is inseparable from the response and public perception that develops on social media. This study aims to classify public sentiment towards the Rp200 trillion fund stimulus policy to Bank HIMBARA based on Instagram user comments and test the performance of the Naïve Bayes Classifier method in analyzing public policy sentiment. This study uses a quantitative approach with text mining and machine learning methods. Data in the form of 1.309 Instagram comments was collected through web scraping techniques from several online media accounts, then processed through text preprocessing and manual labeling stages into positive, neutral, and negative sentiments. Feature weighting was carried out using TF-IDF, then the data were classified using Multinomial Naïve Bayes and Complement Naïve Bayes. The results show that the Complement Naïve Bayes model achieved the best performance with an accuracy of 84%, an F1-score of 81%, and a high ROC-AUC value. These findings indicate that the majority of public sentiment toward the stimulus policy tends to be positive, and that the Naïve Bayes method is effective for social media–based sentiment analysis.

DEVELOPMENT OF WEBSITE-BASED LEARNING MEDIA ON MEDIA ELEMENTS AND TELECOMMUNICATION NETWORKS AT SMK NEGERI 1 PAINAN

2026-02-09T22:40:47+00:00

This study aims to develop a website-based learning media for the Basic Computer and Telecommunication Network Engineering subject, particularly the Media and Telecommunication Networks topic for Grade X TJKT students at SMK Negeri 1 Painan. The research was motivated by limited practical facilities and the lack of interactive learning media, which resulted in low student learning outcomes. This study employed the SDLC iterative model consisting of requirements analysis, design, development, testing, and implementation. Data were collected through expert validation sheets and practicality questionnaires for teachers and students. The developed media integrates visual materials, instructional videos, and interactive quizzes to support independent learning. Validation results indicate that the media achieved a software quality evaluation score of 88.19%, while practicality scores reached 95.48% from teachers and 86.75% from students, categorized as highly practical. These findings demonstrate that the proposed web-based learning media is feasible and effective in supporting the teaching and learning process in vocational education, particularly in improving students’ understanding of abstract networking concepts

APPLICATION OF THE ENTROPY–MARCOS METHOD IN A DECISION SUPPORT SYSTEM FOR SELECTING TOURIST DESTINATIONS IN SEMARANG

2026-02-26T06:20:46+00:00

Tourism is an important sector that supports regional economic growth. In the Semarang area, the increasing number of tourist destinations provides many options for visitors, but it can also make it difficult for tourists to determine the most appropriate destination based on several considerations. Therefore, this study aims to develop a decision support system for selecting tourist destinations using a multi-criteria decision-making approach. The criteria used in this study include travel time, ticket price, tourist facilities, and destination rating. The data were obtained from digital tourism information and questionnaire responses from 64 respondents. The Entropy method was used to determine objective criteria weights based on data variation, while the MARCOS method was applied to rank tourist destination alternatives based on their proximity to ideal and anti-ideal solutions. The results show that Kota Lama is the most recommended tourist destination with the highest final score, followed by Lawang Sewu and Pagoda Avalokitesvara. The validation results using Spearman correlation analysis produced a coefficient value of 0.9879, indicating a very strong agreement between the ranking results generated by the Entropy–MARCOS model and tourist preference rankings. This study contributes by integrating objective weighting and ranking methods using digital tourism data to provide a structured approach for tourism decision support systems.

DEVELOPMENT OF AN ECONOMIC GROWTH DATA VISUALIZATION DASHBOARD FOR PALEMBANG CITY USING THE AGILE METHOD

2026-02-25T01:53:28+00:00

The presentation of economic growth data for Palembang City on the official website of Badan Pusat Statistik (BPS) still faces challenges in terms of usability, navigation, and data visualization. The complex menu structure, poorly organized data presentation, and limited interactive features reduce efficiency in accessing and interpreting economic information. To date, there is no dedicated interactive dashboard that centrally integrates and visualizes Gross Regional Domestic Product (GRDP) data for Palembang City in a user-oriented manner, creating a gap in the provision of accessible regional economic analysis tools. This study aims to develop an Economic Growth Data Visualization Dashboard for Palembang City to present GRDP data in a clearer, more interactive, and user-friendly format. The system was developed using the Agile Development method, consisting of planning, design, development, testing, and evaluation stages. The dashboard was built using Next.js as the frontend framework and MySQL as the database management system. It presents GRDP data at current prices (ADHB), constant prices (ADHK), expenditure components, and business sector categories through interactive charts and dynamic tables. Black-Box testing confirmed that all system features functioned properly. Usability testing using the System Usability Scale (SUS) with 50 respondents resulted in a score of 84.8, categorized as Excellent. The system is feasible as a decision-support tool for regional economic data analysis.

DENGUE FEVER CASE PREDICTION MODEL USING LINEAR REGRESSION WITH EXPLANATORY SEQUENTIAL MIXED METHODS APPROACH

2026-02-23T22:27:21+00:00

Dengue Hemorrhagic Fever (DHF) is an infectious disease in Indonesia, including in Sikka Regency, where the number of cases has increased over the past decade. Predicting the number of DHF cases is crucial to support disease prevention and control policies. This study aims to develop a predictive model for the number of dengue fever cases based on building area, population, and population density, moreover to explain other factors that influence the prediction results. The study uses an explanatory sequential mixed methods approach, and the prediction model is developed using simple linear regression and multiple linear regression. Quantitative data were obtained from the Health Office, the Sikka Regency Statistics Office, and Google Earth; while qualitative data were obtained through interviews with surveillance personnel from the Health Office and several community health centers in the study area, using a purposive sampling technique. The results show that the building area has a weak relationship with the number of DHF cases (R² = 0.10334 for Alok Timur sub-district and R²=0.38055 for Waiblama). After adding the population and population density variables, the R² in Alok Timur increases to 0.46974; and R²=0.41024 for Waiblama; however, the accuracy is still low. The interviews results show that community behavior is the dominant factors of DHF cases. This study indicates that predictive models based on physical environmental and population variables are unable to accurately depict the complexity of dengue fever case distribution. Therefore, the development of models that integrate community behavioral factors is necessary to provide more accurate predictions.

APPLICATION OF THE FIRST COME FIRST SERVED METHOD IN A WEB-BASED MARRIAGE REGISTRATION SYSTEM

2026-03-08T23:02:48+00:00

Marriage registration at the Office of Religious Affairs (KUA) is still largely carried out using conventional methods, which gives rise to various problems such as document accumulation, the risk of file loss, data recording errors, and delays in the verification process, particularly for prospective couples who live far from the KUA office. This process requires prospective couples to visit the KUA office in person and submit physical documents, making it inefficient and ineffective. This study aims to develop a website-based marriage registration system that facilitates online registration for prospective couples and regulates the service order in an orderly and fair. This study applies the First Come First Served (FCFS) method to regulate the order of marriage registration based on the time of registration, ensuring a fair and orderly service process. Furthermore, the system development is carried out using the Waterfall approach, which includes the stages of requirements analysis, system design, implementation, testing, and maintenance. The results show that the website-based marriage registration system operates according to the designed workflow. Alpha testing indicates that all main features function as specified, while beta testing obtained an average score of 4.18 with a percentage of 83.6%, which falls into the good category. These results indicate that the system is well accepted by users and suitable for implementation at the Office of Religious Affairs (KUA).

IMAGE SEGMENTATION OF YOGYAKARTA BATIK PATTERN USING SEGNET

2026-03-12T02:09:04+00:00

Batik is an Indonesian intangible cultural heritage with high artistic value. However, the complexity of classical Yogyakarta patterns, particularly Parang and Kawung, characterized by intricate structures, color variations, and indistinct boundaries, poses significant challenges for automated image processing. Therefore, image segmentation becomes a crucial step in batik identification and digitalization. This study aims to develop an efficient segmentation model for Yogyakarta batik patterns using a modified SegNet architecture. The dataset comprises 720 RGB images, consisting of 360 Parang pattern images and 360 Kawung pattern images. All images were processed into binary ground truth masks through a combination of K-Means Clustering and morphological operations. The SegNet architecture was modified into three encoder and decoder blocks, employing Conv2DTranspose for upsampling and a sigmoid activation function in the output layer. The model was trained for 50 epochs using the Adam optimizer and binary cross entropy loss function. Based on evaluation on the test dataset, the modified SegNet model achieved strong performance with an accuracy of 91.72%, a mean Intersection over Union of 77.23%, and a mean Dice Coefficient of 87.07%. Visual inspection of the prediction results further confirms the model’s ability to accurately separate pattern regions from the background. These findings demonstrate that the modified SegNet architecture performs well in segmenting Parang and Kawung batik patterns and shows strong potential for supporting future batik recognition and digitalization systems.