Jurnal ELTIKOM
Not a member yet
154 research outputs found
Sort by
No-Reference Video Quality Assessment based on The Dover Framework using A Transfer Learning Method
Full text linkNo-reference Video Quality Assessment (VQA) presents a critical challenge in digital multimedia. This study explores video quality measurement using the DOVER framework combined with a transfer learning method. While existing approaches often rely on end-to-end fine-tuning that requires substantial computational resources, this study introduces and validates a more efficient implementation. The model was built using Google Colab and Python, with the KoNViD-1k dataset as the training base. A head-only transfer learning approach was employed, using the DOVER framework as its foundation. This approach addresses a key research gap in resource-efficient no-reference VQA, as many state-of-the-art models remain impractical for real-world deployment due to high computational demands. The training process was conducted over 10 epochs with resource efficiency in mind. The head-only transfer learning technique allows for GPU memory optimization, showing minimal accuracy differences (1%–2%) compared to full end-to-end fine-tuning. Unlike previous studies that compromise performance for efficiency, this approach maintains competitive accuracy while significantly lowering computational costs. The results show that the proposed method delivers accurate and efficient video quality assessments, confirming the potential of the DOVER framework in no-reference VQA. This study highlights a practical balance between computational efficiency and assessment accuracy using transfer learning techniques
Multi-Label Classification for Opinion Mining in The Presidential Election using TF-IDF with NB And SVM
Full text linkPublic opinion plays a crucial role in presidential elections, shaping voter choices and influencing outcomes. Most sentiment analysis studies focus on binary (positive vs. negative) or multiclass (positive, negative, neutral) classification, which limits their ability to capture opinions that express multiple sentiments simultaneously. In presidential elections, a single opinion may support one candidate while criticizing another. This study proposes a MultiLabelBinarizer model to classify candidate and sentiment labels simultaneously—an approach that remains underexplored. The model combines Naïve Bayes (NB) and Support Vector Machine (SVM) for opinion mining using public data and TF-IDF for feature extraction, applying Multinomial and Linear kernels. Performance is evaluated using Accuracy, Precision, Recall, and F1-score. The study is conducted in two stages: developing a multi-label analysis model for presidential candidates and testing the effectiveness of cross-validation. Results show that multi-label classification is effective for both candidate and sentiment categories. Cross-validation with NB and SVM yields high accuracy. NB achieves 0.89 for candidate labels and 0.86 for sentiment labels. SVM performs better, with 0.93 for candidate labels and 0.94 for sentiment labels. While SVM provides higher accuracy, NB offers faster implementation with still competitive results
Prediction of Telkomsel 4G LTE Card Sales using The K-Nearest Neighbor Algorithm
Full text linkAccurate sales prediction is a critical challenge in business decision-making, as factors such as data imbalance, outliers, and overfitting may compromise the reliability of predictive models. This study aims to develop a precise model for predicting card sales using the K-Nearest Neighbor (KNN) algorithm and to offer recommendations for improving prediction quality by addressing issues related to data imbalance and overfitting. The KNN algorithm is applied to analyze a card sales dataset, with preprocessing steps that include detecting missing values, handling outliers, and converting the target attribute into a categorical format. The optimal value of k is identified using the elbow method to determine the model\u27s best accuracy. Findings indicate that the KNN model with k = 1 achieves 100% accuracy, though it shows signs of overfitting, which may hinder its generalizability to new data. Handling outliers and transforming data contributed to improving the model\u27s performance. However, to enhance robustness, further testing with different k values and the use of cross-validation are recommended. Moreover, balancing the dataset and incorporating external variables such as promotional activities or market trends could support more reliable future predictions
A VGG16 CNN-based Method for Multiclass Lung Cancer Classification using CT Imaging
No full textLung cancer is the leading cause of death worldwide among all types of cancer. Early detection and accurate classification are essential to prevent disease progression and improve patient survival rates. One effective approach is the use of computer-aided diagnosis (CAD) systems based on medical imaging, particularly CT scans, which provide high-resolution and non-invasive visualization of lung structures, including blood vessels, soft tissues, and lesions or nodules. This study proposes a VGG16 CNN-based multiclass classification method for lung cancer. Unlike previous studies that primarily focus on binary classification, this research addresses four distinct classes of lung nodule CT images to better reflect complex clinical needs. The modified VGG16 architecture incorporates additional layers, including Flatten, Dense, and Dropout, along with the Softmax activation function, effectively improving classification performance and reducing overfitting risk. An ablation experiment was also conducted by replacing ReLU with LeakyReLU to address the potential “dying ReLU” issue. However, the results indicated that LeakyReLU did not provide significant improvement over the standard ReLU. The proposed model achieved an accuracy of 90.72%, precision of 91.5%, sensitivity of 89.25%, specificity of 96.76%, F1-score of 90%, and a low loss value of 0.37. Furthermore, the modified VGG16-CNN outperformed other CNN architectures, including ResNet50, EfficientNetB1, MobileNetV2, and AlexNet, in multiclass lung cancer image classification. The results demonstrate that the proposed method is effective for diagnosing lung nodules from CT scans and has the potential to support medical professionals in making accurate and timely diagnoses
K-Nearest Neighbor Algorithm for Intelligent Monitoring and Control System Integration in Renewable Energy Applications
No full textA real-time biogas monitoring and control system was developed by integrating the K-Nearest Neighbor (KNN) algorithm into an IoT-based framework for methane pressure prediction and automated control. The system uses an ESP32 microcontroller connected to temperature, gas, and pressure sensors (DHT22, MQ-4, MPX5700) to continuously collect data, with cloud connectivity provided through Firebase and Blynk platforms. The predictive model operates within a live feedback loop, allowing immediate actuation based on forecasted methane conditions. With an optimal parameter of k=4, the KNN model achieved 93.33% accuracy, supported by a mean absolute error (MAE) of 0.18 and a root mean square error (RMSE) of 0.21. A comparative evaluation with Random Forest and Gradient Boosting algorithms showed that, although these models yielded slightly higher accuracy, KNN provided superior computational effi-ciency for embedded deployment. The system maintained stable operation during tests involving sensor anomalies, network interruptions, and data noise. However, redundancy mechanisms and improved vali-dation strategies are recommended to enhance robustness. The findings demonstrate that methane pro-duction can be effectively predicted using temperature and pressure data, with further accuracy im-provements possible through additional process variables such as pH and fermentation age
The Synergy of Blockchain and Cybersecurity: Building Trust in Digital Environments
No full textThe rapid expansion of digital ecosystems has intensified concerns about data security, privacy, and trust. Blockchain technology, characterized by its decentralized, immutable, and transparent nature, offers a transformative approach to strengthening cybersecurity. This paper examines the synergy between blockchain and cybersecurity, emphasizing how blockchain’s cryptographic foundations, consensus mechanisms, and smart contracts can mitigate cyber threats, enhance authentication, and ensure data integrity. By analyzing emerging trends, challenges, and real-world applications, this study underscores the potential of blockchain to reinforce digital trust and resilience across diverse sectors. The findings contribute to the ongoing discourse on secure digital environments by proposing an integrated framework for blockchain-based cybersecurity solution
Optimization Model for Fake Account Detection on Twitter (X) Social Media using Feature Engineering and Machine Learning Approaches
No full textTwitter (X) has become an important platform for community interaction, but this also creates serious challenges due to the proliferation of fake accounts that can harm users and undermine credibility. Previous studies have proposed detection methods but often lacked forensic analysis based on extracted feature information. This study utilizes labeled datasets and supervised evaluation metrics (precision, recall, and F1-score) to validate model performance. Extracting behavioral information from features is crucial for achieving accurate and reliable detection results. The study introduces a novelty in the form of engineered behavioral features that significantly enhance detection accuracy, achieving up to 99.94% using AdaBoost. The proposed approach detects fake accounts on Twitter (X) by extracting key feature information and developing an optimal detection method through machine learning algorithms, including Random Forest, SVM, and AdaBoost. Furthermore, the model is optimized using feature engineering techniques. The novelty of this work lies in the development of engineered features through distribution analysis based on data characteristics and the improvement of classification performance through feature engineering optimization. The initial experiment without feature engineering shows that Random Forest achieved the highest accuracy of 98.77%, followed by AdaBoost at 98.57% and SVM at 95.90%. After applying feature engineering, performance improved, with AdaBoost reaching 99.94%, Random Forest 99.69%, and SVM 99.32%. The proposed model can assist system analysts in detecting fake accounts and contribute to solving forensic cybercrime challenges, particularly in identifying fake social media profiles
IoT-MQTT Protocol-Based Water Sensor System to Monitor Citarum River Water Quality using Arduino Uno R4 Wifi
Full text linkRiver water quality is critical for sustaining life, necessitating advanced monitoring technologies. This study presents a novel IoT-based water monitoring system using the Arduino Uno R4 WiFi and the MQTT protocol, offering significant improvements in real-time data acquisition, reliability, and accessibility. Unlike conventional systems, this approach uniquely integrates advanced microcontroller capabilities and efficient data transmission to address limitations in accuracy and usability in water quality monitoring. The system measures key indicators, including pH, temperature, total dissolved solids (TDS), and turbidity, and provides real-time updates via a solar-powered web interface. Using an exploratory sequential design, the study developed, calibrated, and tested the system, achieving high accuracy with relative errors of 2.50% for pH, 4.15% for temperature, 4.73% for TDS, and 3.08% for turbidity. Feedback from 59 residents near the Citarum River underscores the system\u27s effectiveness and societal relevance, highlighting its potential to enhance public health, support sustainable environmental management, and set a new standard in water monitoring technology
Explainable AI-Driven Convolution Neural Network for Quality Grading of Soybean Seeds
No full textThis study developed a soybean seed grading system based on Explainable Artificial Intelligence (XAI). Traditional soybean quality assessment is time-consuming, and limited research has applied explainable AI methods to the grading process. To address these issues, this study employed classification and XAI methods through several stages. First, it examined five main categories of soybean seed characteristics: broken, immature, intact, skin-damaged, and spotted. Second, it used the Soybean Seeds Dataset contain-ing 5,513 images. Third, data preprocessing was carried out, including image normalization and data division for training and testing. Finally, a Convolutional Neural Network (CNN) model based on the VGG-16 architecture was used for classification experiments. Three XAI methods, namely Shapley Additive Explanations (SHAP), Local Interpretable Model Agnostic Explanations (LIME), and Layerwise Relevance Propagation (LRP), were applied to evaluate model performance and interpretability. The VGG-16 model achieved an accuracy of 91%, with precision, recall, and F1-score values of 0.91, 0.91, and 0.90, respectively. The interpretability analysis using SHAP, LIME, and LRP showed that the model consistently identified key features such as seed shape and surface texture, demonstrating that the system is transparent and reliable in determining soybean seed quality
Integration of PM1200 and IoT for Electrical Energy Monitoring with Web-Based Map Visualization
Full text linkThis study aims to integrate the PM1200 device with Internet of Things (IoT) technology using the Modbus protocol to enable real-time monitoring of electrical energy. The current challenge lies in the limited flexibility of energy monitoring, which is typically restricted to local access and lacks map-based visualizations. To address this, the system integrates interactive maps to provide a clearer and more comprehensive view of energy distribution across different locations. This study seeks to offer an effective energy monitoring solution with data visualized through maps on an interactive web platform. The methodology includes reading data from the PM1200 device via the Modbus protocol, transmitting it to an IoT platform using the MQTT protocol, and displaying the data as maps on a web interface. The findings are expected to support effective energy monitoring and enhance energy management efficiency