Jurnal ELTIKOM
Not a member yet
154 research outputs found
Sort by
Deep Learning Approach for Pneumonia Prediction from X-Rays using A Pretrained Densenet Model
Full text linkPneumonia remains a major global health concern, particularly affecting young children and older adults, contributing to significant morbidity and mortality. Traditional diagnostic methods using chest CT scans are time-consuming and prone to errors due to the reliance on manual interpretation. This study investigates the application of DenseNet architectures DenseNet121, DenseNet169, and DenseNet201—for automated pneumonia detection from chest X-ray images. The dataset, obtained from the Guangzhou Women and Children’s Medical Center, consists of 5,216 training images and 624 testing images categorized into normal and pneumonia cases. Data augmentation techniques, including rotation, normalization, and shear, were applied to improve training efficiency. The DenseNet models were pre-trained on ImageNet and fine-tuned by adding fully connected layers with 256 neurons and sigmoid activation. The models were trained for 20 epochs using the Adam optimizer and binary cross-entropy loss function. Performance evaluation revealed that DenseNet201 outperformed the other models, achieving a precision of 0.99 and a recall of 0.61 for normal cases (F1-score of 0.75) and a precision of 0.81 with a recall of 0.99 for pneumonia cases (F1-score of 0.89). These findings demonstrate that DenseNet201 provides a reliable and effective solution for automated pneumonia detection, offering improved diagnostic efficiency and accuracy compared to traditional methods
Application of Natural Language Processing and LSTM in A Travel Chatbot for Medan City
Full text linkThe tourism sector plays a vital role in economic growth and regional development. Medan, a major city in North Sumatra, offers rich religious, historical, and cultural attractions. However, fragmented and inconsistent information presents challenges for both tourists and destination managers, often complicating travel planning. To address this issue, this study proposes the development of an AI-based chatbot aimed at enhancing the tourism experience in Medan. By integrating Natural Language Processing (NLP) and Long Short-Term Memory (LSTM), the chatbot is designed to deliver accurate, contextual, and conversational responses tailored to users\u27 tourism-related queries. It was trained on a comprehensive dataset compiled from various sources concerning Medan’s tourism. The training ran over 100 epochs, achieving an accuracy of 84.31% and a loss of 0.7594. Validation testing yielded an accuracy of 77.14% and a loss of 2.4233, indicating good generalization to unseen data. End-to-end testing with 312 queries covering all defined intents resulted in a testing accuracy of 75.64%, confirming the model’s practical effectiveness. The findings demonstrate that the chatbot can accurately interpret user input, classify information, and enhance user interaction. supports the digital transformation of Medan’s tourism sector by introducing a reliable, AI-driven tool for seamless travel planning and engagement
Design and Performance Evaluation of A Portable Low-Head Pico-Hydro System using A Rewound Axial Generator for Rural Energy
Full text linkThis study evaluates the performance of a pico-hydro system installed on a river with low head and discharge. The system was assessed under no-load and varying load conditions (25–100%). The results indicate that the generator performs according to the initial design, despite some fluctuations in output parameters. Under no-load conditions, the generator maintains a stable output voltage between 12–14 VAC, with a rotational speed of 590–600 RPM, a system frequency of 59–60 Hz, and zero current. The step-up transformer successfully raises the voltage to 220–222 V with high stability, although minor ripple is observed in the output signal. Under load, the generator voltage slightly decreases to 12–14 V as the load increases. The rotational speed also declines (560–590 RPM), affecting frequency stability, which drops from 59 Hz at 25% load to 56 Hz at full load. The current rises proportionally with the load, from 0.10 A at 25% to 0.45 A at 100%. The observed performance drop under load highlights the effect of load on generator speed and overall system output. The primary impacts of the 25–100% load range are evident in generator speed, frequency stability, and waveform quality. Overall, the system performs satisfactorily for low-head pico-hydro applications with a power capacity of up to 100 Watts, suitable for rural street lighting
Efficient Strategy for Distribution Transformer Replacement: A Study on Replacement Methods in Power Systems
No full textA dependable and efficient electric power distribution system is increasingly required due to growing industrial and population demand. Many distribution transformers currently in use are outdated, which increases operating expenses and reduces efficiency. This study examines the issue of determining the most cost-effective time to replace aging distribution transformers to optimize operational expenses and enhance service performance. The purpose of this study is to calculate the annual operating costs of dis-tribution transformers at PT PLN (Persero) Makassar Branch and to identify the optimal replacement time to support more effective decision-making. Using an economic feasibility analysis based on the annuity method and linear regression, this study compares the annual operating costs of old transformers with those of new ones. The findings indicate that the annual cost of old transformers is higher than that of new ones, suggesting that replacing old transformers is more economical. The results show that operating expenses can be reduced and distribution efficiency improved through a systematic transformer replace-ment approach. Based on economic engineering analysis, this study provides a practical and relevant model for transformer replacement decision-making that can assist asset management and investment planning in the power industry
Smart Hydroponic Nutrient Monitoring and Control System using Fuzzy Logic and IoT
No full textHydroponic farming offers an efficient and sustainable solution for modern agriculture, although main-taining stable nutrient levels remains a key challenge. Previous systems often exhibited high overshoot and were unable to adapt to external changes or disturbances, and no existing research has adaptively regulated nutrient levels based on the plant’s growth stage. Therefore, this study aims to develop a smart nutrient monitoring and control system for hydroponics using a Sugeno-type fuzzy logic controller inte-grated with an IoT-based application. Unlike prior systems that rely on fixed setpoints or only address nutrient deficiency, this system dynamically adjusts nutrient and water levels based on real time sensor data and plant growth phase. The system utilizes nutrient, water level, and temperature sensors connected to an ESP32 microcontroller, with fuzzy logic determining solenoid valve activation duration. The control system achieved stable regulation with zero overshoot, a settling time of 840 seconds, and effective recov-ery from nutrient disturbances. Growth tests on celery showed a 102.6% improvement in height, 275% in stem diameter, and 112.5% in leaf width compared to manual control. IoT monitoring via a mobile appli-cation ensured real time visibility of hydroponic parameters. These results demonstrate the system’s ca-pability to maintain optimal nutrient levels, improve control precision, and enhance plant productivity
Fine-Grained Plant Classification using Vision Transformers with Optimized MLP Heads
No full textAutomatic plant species classification is crucial for advancing education and biodiversity conserva-tion. Deep learning models, such as Vision Transformer (ViT), have demonstrated strong performance in plant species classification tasks. However, limited research explored the impact of hyperparameters in the Multi-Layer Perceptron (MLP) head of ViT models for plant-species classification. This study investi-gated the influence of learning rates, number of neurons, and activation functions on model performance. It also evaluated efficiency in both CPU and GPU environments. The objective was to determine the opti-mal configuration by analyzing accuracy, F1-score, and computation time. Two ViT models, ViT-B/16 and ViT-L/16, were tested using the VNPlant-200 dataset, which contains 200 plant species. Thirteen activa-tion functions, multiple learning rates, and neuron configurations were examined. The results showed that the Tanh activation function, combined with a learning rate of 10-4 and 1024 neurons, yielded the best performance on the ViT-B/16 model, achieving an accuracy of 0.9692 and F1-score of 0.9684. Meanwhile, the Hard Tanh activation function, with a learning rate of 10-4 and 256 neurons, delivered the best results on the ViT-L/16 model, achieving an accuracy of 0.9855 and an F1-score of 0.9854. Computational analy-sis showed that ViT-B/16 achieved an average inference time of 0.0159 seconds on a GPU and 0.8902 seconds on a CPU, while ViT-L/16 took 0.0492 seconds on a GPU and 2.8335 seconds on a CPU. These findings highlight the importance of selecting suitable activation functions, learning rates, and neuron configurations to optimize model performance while maintaining computational efficiency
Classification of Diabetes Mellitus using Decision Trees
No full textDiabetes Mellitus is a global health concern, with its prevalence and incidence rising sharply world-wide, including in Indonesia. Several factors contribute to the onset of diabetes mellitus, such as heredity, age, weight, and blood pressure. Managing blood sugar levels, maintaining a balanced diet, exercising regularly, and undergoing early screening when necessary are among the key measures to prevent and control this disease. Early diagnosis is essential to reduce both the number of cases and the associated risks. This study aims to detect diabetes mellitus using classification techniques. The method involves several subprocesses within the classification procedure. The first stage, data preprocessing, includes feature selection and data cleaning. The resulting preprocessed data are then used in the classification stage, specifically the learning subprocess, to generate a decision tree model. Model construction employs pruning, followed by training and performance evaluation. The study utilizes a diabetes dataset obtained from kaggle.com, consisting of 768 records. The dataset includes attributes such as Pregnancies, Glucose, Blood Pressure, Skin Thickness, Insulin, Body Mass Index (BMI), Diabetes Pedigree Function, Age, and the label Outcome. Testing was conducted using decision trees with maximum depths of 3, 5, 7, 10, and 15. The results show that the highest accuracy (88.56%) occurred at a maximum depth of 5, while the highest recall (100%) was achieved at a depth of 3. The highest precision (47.37%) and specificity (95.85%) were also obtained at a depth of 3
Wound Depth Measurement System in Forensic Cases using Image Processing and Machine Learning
No full textAccurate evaluation of wound depth is crucial in forensic investigations, as it significantly affects case assessments and outcomes. This study introduces a method for classifying wound depth using a Support Vector Machine (SVM) model and compares its performance with Decision Tree and Logistic Regression models. The classification was based on color features extracted from HSV and LAB color spaces. The da-taset consisted of 76 images categorized into three stages: stage 2 (36 images), stage 3 (12 images), and stage 4 (28 images). Model performance was evaluated using confusion matrices, precision, recall, and F1-score. The SVM model achieved an overall accuracy of 85%, demonstrating higher precision and re-call across all stages compared to the Decision Tree and Logistic Regression models, which achieved 50% and 70%, respectively. The results indicate that the SVM model performed particularly well in distinguish-ing stage 2 wounds, although differentiating between stages 3 and 4 remained challenging. Overall, the proposed system shows potential to enhance the accuracy and efficiency of forensic wound evaluation by providing a rapid and objective classification tool. However, as the system was tested on a limited dataset under controlled conditions, further research should expand the dataset, incorporate additional features, and explore other machine learning algorithms to improve robustness and applicability in real forensic contexts
An Intelligent Fuzzy Logic-Controlled IoT System for Efficient Hydroponic Plant Monitoring and Automation
Full text linkThis paper addresses the challenges of optimizing environmental conditions in hydroponic farming by integrating an Intelligent Fuzzy Logic-Controlled IoT System. The research problem lies in the inefficiency of traditional hydroponic monitoring systems, particularly in maintaining ideal conditions for plant growth while minimizing resource waste. This study aims to develop a system that leverages IoT technology and fuzzy logic to monitor and automate hydroponic processes more efficiently. Using sensors, the system continuously tracks key environmental parameters such as temperature, humidity, soil moisture, pH levels, and total dissolved solids (TDS). A fuzzy logic controller (FLC) triggers actions based on predefined rules. During testing, the system showed effective performance—for example, activating fans when temperature (31.2°C) and humidity (60%) indicated a need for cooling, and adjusting nutrient levels when pH (5.8) and TDS (450 ppm) were suboptimal. The system offers practical benefits through real-time adaptation using defuzzification and aggregation, ensuring precise resource control, improving efficiency, and reducing waste. This study highlights the system\u27s potential to support sustainable agriculture by providing scalable solutions that enhance plant growth and optimize resource use, especially for small-scale farmers and urban farming initiatives
Imbalanced Text Classification on Tourism Reviews using Ada-boost Naïve Bayes
Full text linkHidden paradise is a term that aptly describes the island of Madura, which offers diverse tourism potential. Through the Google Maps application, tourists can access sentiment-based information about various attractions in Madura, serving both as a reference before visiting and as evaluation material for the local government. The Multinomial Naïve Bayes method is used for text classification due to its simplicity and effectiveness in handling text mining tasks. The sentiment classification is divided into three categories: positive, negative, and mixed. Initial analysis revealed an imbalance in sentiment data, with most reviews being positive. To address this, sampling techniques—both oversampling and undersampling—were applied to achieve a more balanced data distribution. Additionally, the Adaptive Boosting ensemble method was used to enhance the accuracy of the Multinomial Naïve Bayes model. The dataset was split into training and testing sets using ratios of 60:40, 70:30, and 80:20 to evaluate the model’s stability and reliability. The results showed that the highest F1-score, 84.1%, was achieved using the Multinomial Naïve Bayes method with Adaptive Boosting, which outperformed the model without boosting, which had an accuracy of 76%