Jurnal Riset Informatika

EMPLOYEE PERFORMANCE ASSESSMENT BASED ON MONTHLY PERFORMANCE USING AHP-SAW AT UPPKH PAMEKASAN REGENCY

2026-04-02T02:04:04+00:00

Employee performance assessment plays an important role in improving organizational productivity and supporting decision-making processes. However, the evaluation process at UPPKH Pamekasan Regency is still conducted manually, which often leads to subjectivity, inconsistency, and inefficiency. This study aims to develop a Decision Support System (DSS) for employee performance assessment using the combination of Analytical Hierarchy Process (AHP) and Simple Additive Weighting (SAW) methods. The AHP method is used to determine the weight of nine evaluation criteria through pairwise comparison, while the SAW method is applied to rank 20 PKH facilitators based on their performance scores. The system is implemented as a web-based application using the Laravel framework and MySQL database. The results show that the system is able to produce objective and structured rankings, where the highest preference value is obtained by alternative K4 with a score of 0.922. Furthermore, the accuracy of the method is evaluated using the Spearman Rank Correlation test, resulting in a coefficient of 0.97143, which indicates a very strong correlation between the system-generated rankings and the manual rankings from UPPKH. In addition, black box testing confirms that all system functionalities operate correctly. Therefore, the proposed system is effective in reducing subjectivity, improving efficiency, and supporting accurate decision-making in employee performance evaluation.

HYBRID ESP-NOW AND MQTT-BASED MONITORING AND EARLY WARNING SYSTEM FOR RRI MANADO TRANSMITTER ROOM

2026-04-06T22:38:58+00:00

The transmitter room of LPP RRI Manado is a vital operational center housing high-power electronic equipment susceptible to damage from temperature fluctuations, humidity instability, abnormal machine noise, and fire hazards. Currently, no automated monitoring system has been implemented to continuously observe environmental conditions in real-time, resulting in delayed detection of technical anomalies. This research aims to design and implement an IoT-based monitoring and early warning system for the transmitter room environment. The novelty of this system lies in its hybrid ESP-NOW and MQTT communication architecture, specifically tailored for the multi-parameter environmental monitoring demands of broadcast transmitter facilities, where neither protocol alone could fulfill both low-latency local transmission and real-time remote dashboard access simultaneously. The system employs ESP-NOW for wireless inter-building data transmission and MQTT for real-time integration to a web-based monitoring dashboard. Environmental parameters were selected based on the primary physical risk factors in high-power electronic environments: temperature and humidity (DHT22) to detect thermal and moisture anomalies, sound intensity (GY-MAX4466) to identify mechanical failure indicators, and smoke concentration (MQ-2) as an early fire hazard indicator. The research method used is quantitative with an experimental approach, whereby numerical data were collected from sensor readings and communication performance tests, then analyzed to evaluate system accuracy, latency, and packet loss. The expected outcome is a system that enables centralized monitoring for TMB technicians and delivers early warning notifications when abnormal conditions are detected, ensuring operational reliability and continuity of LPP RRI Manado broadcast services.

A HOLISTIC AI-DRIVEN ENERGY-EFFICIENT IOT FRAMEWORK FOR SMART AGRICULTURE USING MULTI-RESOURCE OPTIMIZATION

2026-06-18T05:58:46+00:00

The rapid adoption of Internet of Things (IoT) technologies has accelerated the development of smart agriculture systems. However, existing studies predominantly focus on single-resource optimization and lack integrated artificial intelligence (AI) approaches within distributed architectures, resulting in suboptimal system-wide performance. This study proposes an AI-driven energy-efficient IoT framework that integrates the Random Forest algorithm with an edge–fog–cloud computing architecture to enable holistic multi-resource optimization. A quantitative simulation-based approach was employed using soil moisture data from the NASA SMAP dataset, with a case study in Magelang, Indonesia. The system was evaluated using key performance metrics, including energy consumption, network latency, packet delivery ratio (PDR), and water usage efficiency. The results demonstrate significant improvements, including a 28.65% reduction in energy consumption, a 31.43% decrease in latency, an increase in PDR to 96.8%, and a 20.3% improvement in water usage efficiency. Statistical validation confirms that these improvements are significant (p < 0.05). The main contribution of this study lies in the development of a holistic AI-driven IoT framework that simultaneously optimizes energy, water, computation, and communication without trade-offs. The proposed approach offers a scalable, adaptive, and efficient solution for real-world smart agriculture systems.

A COMPARATIVE STUDY OF DISTANCE METRICS AND NEIGHBOR SELECTION IN K-NEAREST NEIGHBOR FOR VOCATIONAL STUDENT PERFORMANCE CLASSIFICATION

2026-04-23T23:20:25+00:00

This study aims to evaluate parameter sensitivity in the K-Nearest Neighbor (KNN) algorithm, particularly the selection of distance metrics and k-values, for classifying academic performance in vocational education with heterogeneous and imbalanced data characteristics. The dataset consists of 750 first-year students from the Informatics Management program, including academic attributes (GPA, attendance, and core course grades) and demographic attributes (age, gender, educational background, and economic status). Data preprocessing involves data cleaning, one-hot encoding, Z-score normalization, and handling class imbalance using SMOTE. Model evaluation is conducted using K-Fold Cross Validation with accuracy, precision, recall, and macro-average F1-score as performance metrics. The results show that KNN performance is highly influenced by the combination of distance metrics and k-values. All metrics achieve accuracy above 84%, but differ in handling class imbalance. The Chebyshev metric (k = 10) provides the best balance with an F1-score of 0.6468, while the Minkowski metric (p = 3) achieves the highest recall of 0.7334. The Euclidean metric attains the highest accuracy of 0.8504 (k = 11), but tends to be biased toward the majority class. These findings indicate that optimizing KNN parameters should not rely solely on accuracy, but also consider balanced performance across classes. This study provides a practical evaluation framework for selecting KNN parameters to support more robust and fair academic prediction systems in vocational education data.

STEMMINDO: A WEB-BASED INDONESIAN STEMMING ENGINE USING ENHANCED CONFIX STRIPPING

2026-06-18T05:58:59+00:00

Stemming is an essential preprocessing stage in Natural Language Processing (NLP), particularly for Indonesian, which has complex affixation patterns. Most Indonesian stemming implementations are provided as programming libraries, making them less accessible for beginners, educators, and non-programmer researchers. This study presents Stemmindo, a lightweight web-based Indonesian root word search application implementing the Enhanced Confix Stripping (ECS) algorithm using the Laravel framework. Unlike conventional stemming libraries, the system provides a real-time and modular interface that enables users to explore Indonesian morphological processing without writing program code. The novelty of this research lies in the implementation of ECS within an accessible web-based educational tool. Evaluation was conducted using affixation pattern testing, rule-based testing, and real-text evaluation. Testing on 20 affixation patterns achieved 90% accuracy, while evaluation on 100 words representing 33 derived prefix rules achieved 94% accuracy. After applying failure-handling strategies through exception lists and rule-level accommodations, the accuracy increased to 98%. Real-text evaluation was conducted using 1,742 words collected from Indonesian educational web content. After preprocessing and filtering, 564 unique words were evaluated, of which 366 stemming results were successfully matched with the corpus, while the remaining cases mainly consisted of named entities, noisy input, ambiguous forms, overstemming, and understemming. These findings indicate that the proposed system performs effectively for common Indonesian morphological patterns while remaining practical for educational and experimental NLP usage. Future work includes improving reduplication handling, expanding lexical resources, and enhancing accessibility features.

AGILE IMPLEMENTATION IN MOBILE POINT OF SALE SYSTEM DEVELOPMENT FOR BUSINESS DIGITALIZATION

2026-05-12T11:22:03+00:00

The development of information technology drives the need for a flexible, efficient, and easy-to-use Point of Sale (POS) system to support retail business operations. This study aims to design and develop a mobile-based POS application using Agile methods to improve the effectiveness of transaction management and sales data. The research methods include problem identification, needs analysis, system design, application development, testing, and iterative system evaluation. The system design was carried out using use case diagrams and Entity Relationship Diagrams (ERD), while the implementation was developed on a mobile platform with key features including user authentication, product management, sales transactions, stock management, reports, and owner and cashier access rights settings. Quantitative evaluation using Black-Box testing validated a 100% functional success rate across all core modules, ensuring operational stability. Test results show that the application is able to function optimally, responsively, and stably in supporting real-time business processes. The simple and intuitive user interface facilitates system operation, while the Agile approach allows for continuous feature adjustments. Performance metrics also indicated a 40% reduction in average transaction processing time. Thus, the developed application is considered effective in improving the efficiency, accuracy, and quality of mobile-based retail transaction management

COMPARATIVE PERFORMANCE OF EFFICIENTNET-B0 AND RESNET-50 FOR MELANOMA DETECTION IN DERMOSCOPY IMAGES

2026-06-18T05:58:32+00:00

Melanoma is the most aggressive form of skin cancer with high metastatic potential, and early detection is crucial for improving patient survival. Although deep learning models such as ResNet-50 and EfficientNet-B0 have shown promising results in melanoma classification, systematic comparisons using identical experimental protocols remain limited. This study aims to comprehensively compare the performance of EfficientNet-B0 and ResNet-50 in detecting melanoma from dermoscopy images across multiple evaluation dimensions, including accuracy, precision, recall, F1-score, and computational efficiency. A quantitative experimental research design was employed using the publicly available HAM10000 dataset, consisting of 10,015 dermoscopy images categorized into melanoma and non-melanoma classes. Both models were implemented using transfer learning with ImageNet pretrained weights, trained under identical conditions including data augmentation, class imbalance handling using weighted loss, and standardized hyperparameters. Results showed that EfficientNet-B0 achieved superior performance with 91.5% accuracy, 89.8% precision, 88.2% recall, and 89.0% F1-score, compared to ResNet-50 which achieved 89.2% accuracy, 87.5% precision, 85.3% recall, and 86.4% F1-score. Furthermore, EfficientNet-B0 demonstrated significant computational advantages with only 5.3 million parameters (79% fewer than ResNet-50’s 25.6 million). In conclusion, EfficientNet-B0 outperforms ResNet-50 in both accuracy and computational efficiency, making it more suitable for deployment in resource-constrained clinical environments such as mobile telemedicine applications.

BREAST TUMOR CLASSIFICATION USING RANDOM FOREST WITH FEATURE SELECTION AND GRIDSEARCHCV OPTIMIZATION

2026-06-18T05:58:05+00:00

Breast tumor classification into benign and malignant categories is an important challenge in the medical field because diagnostic errors can lead to delayed treatment or unnecessary medical procedures. This study aims to analyze the performance of Random Forest and evaluate the effects of feature selection and GridSearchCV hyperparameter optimization on breast tumor classification. The study used the Wisconsin Breast Cancer Diagnostic Dataset, consisting of 569 samples with 30 numerical features extracted from Fine Needle Aspiration (FNA) examinations. Four sequential Random Forest model configurations were compared: baseline Random Forest, Random Forest with feature selection, Random Forest with GridSearchCV optimization, and the integration of feature selection with GridSearchCV. Feature selection was performed using feature importance scores with ROC-AUC-based cross-validation to determine the optimal feature subset. Model evaluation was conducted using accuracy, precision, recall, F1-score, ROC-AUC, confusion matrix, and train-test gap. The results showed that all models achieved the same accuracy of 97.37%, precision of 1.0000, recall of 0.9286, and F1-score of 0.9630. However, the integrated model achieved the highest ROC-AUC of 0.9977 with the smallest train-test gap of 0.0241 while reducing the number of features from 30 to 15. These findings indicate that integrating feature selection and GridSearchCV improves model stability, efficiency, and discriminative capability without reducing classification performance, addressing the limitation of prior studies that applied these techniques separately.

DIGITAL IMAGE PROCESSING FOR BRAIN TUMOR CLASSIFICATION IN HUMANS USING CONVOLUTIONAL NEURAL NETWORKS

2026-06-18T05:57:50+00:00

The rapid development of digital technology has encouraged its utilization in various aspects of life, including the medical field. One significant application is digital image processing, which is used to enhance the quality and utility of medical imagery such as MRI and CT scans. This technology is highly relevant in diagnosing brain diseases, particularly brain tumors, which require high precision given the organ's complexity. This research focuses on the classification of brain tumor diseases using MRI images through the Convolutional Neural Network (CNN) method. CNN was selected due to its ability to extract essential features from MRI images, enabling it to identify complex patterns that are difficult for the human eye to recognize. With proper training, the CNN model is capable of distinguishing between healthy brain MRI images and those with tumors with an accuracy of 80%. These results demonstrate great potential in accelerating and improving the accuracy of the diagnostic process, which in turn assists in determining appropriate and effective treatment steps. This study provides a significant contribution to the development of medical diagnostic technology, specifically in brain tumor classification. Through the application of advanced digital image processing technology, it is expected that more efficient and accurate diagnostic tools can be created, thereby improving the quality of healthcare and patient treatment outcomes.

MACHINE LEARNING APPROACH FOR TRANSFORMER CONDITION ASSESSMENT USING K-MEANS CLUSTERING AND MULTI-CLASSIFIER MODELS

2026-06-18T05:57:37+00:00

Transformers play a critical role in power systems, yet their degradation is often difficult to detect due to complex influencing factors. Conventional diagnostic methods, such as Dissolved Gas Analysis (DGA), are time-consuming and rely heavily on expert interpretation. This study proposes a machine learning approach for transformer condition assessment by combining clustering and classification techniques. K-Means clustering is first applied to identify patterns in transformer condition data without prior labeling, with the optimal number of clusters determined as three using the Elbow Method. The resulting clusters are then used as pseudo-labels to train multiple classification models, including KNN, Decision Tree, SVM, Gradient Boosting, Extra Trees, and Voting Classifier. The results show that all models achieve high performance, with accuracy above 94%. Ensemble methods, particularly Gradient Boosting and Voting Classifier, achieve the best performance with an accuracy of 98.30%. These findings demonstrate that the proposed approach effectively improves transformer condition assessment and supports faster and more reliable maintenance decision-making.

SENTIMENT ANALYSIS OF IPUSNAS REVIEWS USING NAIVE BAYES AND K-NEAREST NEIGHBOR ALGORITHMS

2026-06-18T05:58:18+00:00

This study aims to analyze the sentiment of iPusnas application user reviews using the classification method with the K-Nearest Neighbor (KNN) and Naive Bayes algorithms. The data used are secondary data in the form of user reviews obtained from the Google Play Store in the period of January to December 2025 totaling 2415 reviews. This study uses a text mining approach with text preprocessing stages, feature extraction using the Bag of Words (BoW) method, and sentiment clas, sification using the Naive Bayes and K-Nearest Neighbor (KNN) algorithms. Model evaluation uses Test and ScoreConfusion Matrix, and Word Cloud. The results show that the Naive Bayes method has better performance with an accuracy value of 0.705 compared to K-Nearest Neighbor (KNN) with an accuracy value of 0.615. Testing the K parameter in the KNN algorithm shows that the best K value is obtained at K = 6 with an accuracy of 0.615. This study shows that Naive Bayes is more effective in classifying sentiment in iPusnas application user reviews.

TRANSFER LEARNING ARCHITECTURE SELECTION FOR REMOTE SENSING SCENE CLASSIFICATION

2026-06-20T05:58:21+00:00

Selecting a deep learning architecture for classifying remote sensing scenes usually involves comparing published accuracy across papers that each use different training protocols, making it unclear whether accuracy gaps reflect architecture or training differences. We isolate the architecture variable by evaluating eight models from three design families, five classical CNNs (ResNet-50, ResNet-101, DenseNet-121, EfficientNet-B0, EfficientNet-B3), two vision transformers (ViT-B/16, Swin Transformer), and one modernized CNN (ConvNeXt-Tiny), under identical training conditions on EuroSAT (10 classes, 27,000 Sentinel-2 patches) and UC Merced (21 classes, 2,100 aerial photographs). Every model shares the same ImageNet-1K initialization, AdamW optimizer, augmentation pipeline, and early stopping rule. ConvNeXt-Tiny reached the highest accuracy on EuroSAT (99.11%) and Swin-T on UC Merced (99.76%), but the accuracy range on EuroSAT was only 0.41 percentage points (1.66 on UC Merced). McNemar's test confirmed that most pairwise differences were not significant. EfficientNet-B0, the smallest model at 4.0M parameters, reached 98.76% and 99.52% while using 21x fewer parameters than ViT-B/16. On these two well-studied benchmarks, a single uniform training configuration was sufficient to bring all architectures to near-identical performance. This convergence, observed under one fixed protocol and a single data partition, suggests that on saturated classification tasks the choice of architecture may be secondary to the choice of training procedure. Whether this convergence holds on harder benchmarks, under architecture-specific optimal configurations, or with domain-specific pretraining remains to be tested

DETECTION OF SUGARCANE LEAF DISEASES USING MOBILENETV3LARGE-BASED TRANSFER LEARNING FOR MOBILE APPLICATIONS

2026-06-19T23:23:46+00:00

Sugarcane is one of Indonesia's key plantation commodities with a critical role in fulfilling national sugar demand and supporting bioethanol production. However, sugarcane productivity remains low due to leaf diseases that reduce crop quality and yields, while slow or inaccurate identification accelerates their spread. This study proposes and develops a mobile-based sugarcane leaf disease detection system using transfer learning with the MobileNetV3Large architecture to classify 11 disease classes. Two dataset scenarios were applied: Scenario 1 using the SLD Dataset with 6,748 images and Scenario 2 combining the SLD and Sugarcane Smut datasets totaling 14,804 images. Each scenario was trained under three optimizer configurations: Adam, RMSprop, and SGD, to identify the best-performing combination. Results show that Adam achieved the highest validation accuracy in both scenarios, reaching 94.24% in Scenario 1 and 97.43% in Scenario 2, with corresponding test accuracies of 94.91% and 97.31% respectively. The final model was deployed as a Flutter-based mobile application capable of performing real-time disease detection through image upload or camera capture, providing an accessible tool for farmers to identify sugarcane leaf diseases efficiently.

TIME SERIES FORECASTING AND CLASSIFICATION OF POTENTIAL SAFETY RISKS OF STORING RADIOACTIVE WASTE NEAR SURFACE DISPOSAL

2026-06-19T23:23:49+00:00

Long-term radioactive waste management, especially at Near Surface Disposal (NSD) facilities, requires a predictive approach and adaptive monitoring system to anticipate risks to groundwater quality. This research aims to develop a time series model to predict groundwater level parameters including depth, pH, and tds and integrate it with a rule-based ESG risk classification system and machine learning. The method used includes the Prophet time series model for predicting groundwater parameters in the next 50 years. The prediction results are classified using rule-based classification which is then evaluated using the Random Forest algorithm. The final application was developed web-based using Streamlit. The Prophet model provided the best prediction performance for depth MAE: 0.71; MAPE: 7.41% and pH MAE: 0.21; MAPE: 4.89%, but less accurate for TDS MAE: 12.16; MAPE: 31,62%. The Random Forest model produced classification accuracy of up to 98% and was able to replicate the rule-based classification system well. The integration of these models can produce a predictive system that supports decision making in sustainable radioactive waste management.

APPLICATION OF WEB-BASED DECISION SUPPORT SYSTEM FOR SUBSIDIZED 3 KG LPG DISTRIBUTION ROUTE OPTIMIZATION USING CLARKE AND WRIGHT SAVINGS ALGORITHM

2026-06-19T23:23:48+00:00

The distribution of subsidized 3 kg LPG cylinders often relies on manually planned routes based on drivers’ experience, resulting in inefficient distribution operations. This study aims to develop a web-based Decision Support System integrated with WebGIS visualization for optimizing subsidized 3 kg LPG distribution routes using the Clarke and Wright Savings Algorithm. The research method consisted of problem analysis, literature review, data collection, route optimization, system development, and system evaluation. Route optimization was performed using demand data from 22 LPG outlets while considering vehicle capacity constraints. The optimization results generated five distribution routes with a total travel distance of 85.87 km, compared with 98.78 km for the company's existing routes, resulting in a distance reduction of 12.91 km (13.06%). System verification showed that the system produced results identical to manual Clarke and Wright Savings calculations. Blackbox Testing indicated that all system functions operated successfully, while User Acceptance Testing (UAT) achieved a score of 91.8%, indicating a very high level of user acceptance. These results demonstrate that the developed system can support more efficient and systematic LPG distribution planning.

CLIENT-SIDE ONLINE GAMBLING DETECTION USING MULTI-LAYER CASCADE PATTERN MATCHING IN MANIFEST V3 CHROME EXTENSIONS

2026-06-18T05:57:24+00:00

Online gambling sites in Indonesia generated IDR 155.4 trillion in transactions in 2025 with 3.2 million active players, yet DNS filtering the dominant countermeasure blocks only 0.64% of observed gambling traffic. Network-layer approaches fail structurally: they cannot intercept content via VPN, DNS-over-HTTPS, or direct IP access, and they cannot detect the domain neutralization used by the majority of Indonesian gambling operators. This paper proposes GUPI (Gambling URL Pattern Interceptor), a Chrome Extension implementing a three-layer cascade detection architecture running entirely client-side under Manifest V3 without external server dependencies. Layer 1 applies weighted lexical scoring to URL features. Layer 2 applies DOM keyword pattern matching with conditional context suppression. Layer 3 applies CSS selector-based DOM structural heuristic scoring to detect gambling-characteristic page architectures when text-level signals are absent. GUPI was evaluated on 926 URLs (326 gambling, 600 benign) across three sequential configurations. The full system achieves 98.81% accuracy, 99.07% precision, 97.55% recall, 98.30% F1-score, and 0.50% false positive rate.