Jurnal Nasional Teknik Elektro dan Teknologi Informasi

Prototype of Internet of Things-Based Automatic Hydroponic System

Isyara Khairani, Kiki Prawiroredjo — Mon, 20 Jan 2025 10:13:26 +0700

The increase in food needs, including vegetables and fruits, corresponds with population growth. However, agricultural land is increasingly declining due to land conversion. This decline can threaten national food security. Utilizing hydroponic systems for plant cultivation is one of the efforts to adapt to land reduction, land degradations, and adverse impacts of global climate change. Unfortunately, hydroponic cultivation requires constant monitoring of plant nutrition. This research aimed to create an automatic hydroponic system that controlled the process of regulating nutrients to save growers time and energy. Through Internet of things (IoT) technology, automatic hydroponic cultivation can monitor plant life, temperature, humidity, water level in reservoirs, total dissolved solids (TDS), and pH of nutrient solutions. In addition, it can visually monitor plants through Android applications. The hydroponic system used for planting was the nutrient film technique, and the plant cultivated was lettuce. The system consisted of TDS sensors to measure TDS, analog pH sensors to measure the pH, the HC-SR04 ultrasonic sensors to measure the water level in the reservoir, DHT11 sensors, ESP32 microcontrollers, and ESP32-CAM to monitor plant growth remotely. Based on system testing results, the average of TDS increased from 600 ppm in the first week to 900 ppm in the fifth week, the average pH was 6.19, and the average water level in the reservoir was 20.89 cm. All test result parameters are at the designed values.

Digitized Cursive Handwriting for Determining FMS in Early School-Age Children

Nurul Zainal Fanani, Ika Widiastuti, Khamid, Laszlo T. Koczy — Mon, 20 Jan 2025 10:13:45 +0700

Assessing fine motor skills (FMS) in early school-age children is crucial for insights into their school readiness. In many countries, including Indonesia, teachers assess FMS by observing handwriting, often with the aid of an educational psychologist. However, this approach can be subjective and prone to observer bias. This study aimed to classify children’s FMS based on their cursive writing abilities using a digitizer to capture data. The system recorded data in real-time as children wrote in cursive, capturing the stylus’s relative position on the digitizer board (including x, y, and z positions), and pressure values, which served as features in the classification process. The study involved 40 1st and 2nd-grade students from various elementary schools. The data recording process generated substantial raw datasets. The random forest algorithm, renowned for its effectiveness in analyzing large datasets, was employed for classification. The results demonstrated this method’s efficacy in identifying FMS, achieving an accuracy rate of approximately 97.3%. This study concludes that integrating a digitizer with the random forest classification method provides a reliable and objective approach to assessing FMS in children, reducing observer bias, and ensuring precise results. In the long term, this approach can significantly enhance the accuracy of FMS assessments, enabling better-targeted interventions and support for children in need.

Trust Perception and Information Use for Informational Website: Structural Equation Modelling Approach

Umi Proboyekti, Ridi Ferdiana, P. Insap Santosa — Wed, 26 Feb 2025 13:47:07 +0700

Trust is described in various contexts, such as e-commerce, e-government, reviews, and online health information. Credibility and information quality are fundamental to building trust in those contexts. This study aimed to develop trust perception (TP) and information use (IU) indicators in an information evaluation context. Indicators were developed through three processes: searching, grouping, and construction. Relevant indicators were grouped based on similarities to construct statements, which were validated for face and content validity by three experts. The validated TP and IU were then tested using the partial least squares structural equation modeling (PLS)-SEM. The data used for measurement obtained from 110 participants comprising 55 Indonesian academic librarians and 55 university students. Participants responded to indicator statements after evaluating information from four prepared informational websites. This study yielded five TP indicators and a single IU indicator, where TP significantly predicted IU. The five indicators described TP as make-sense information relevant to needs, provided by trusted authors and providers, and accompanied by accessible author information, provider information, and reference sources. IU was described as the information used for its credibility. The measurement demonstrated distinct participant behaviors. Differences in needs influenced assessments, while author and provider trustworthiness showed no bias toward participant type. Trust perception significantly predicted IU, with moderate model fit and varying predictive strengths across the websites. Tested as reliable, valid, and a significant predictor of IU, TP serves as a tool for examining factors that potentially influence trust in online information.

Comparison of KNN and SVM Algorithms Performance Using SMOTE to Classify Diabetes

Asri Mulyani, Sarah Khoerunisa, Dede Kurniadi — Wed, 26 Feb 2025 13:54:05 +0700

Diabetes frequently goes undetected or is diagnosed too late. Consequently, it may lead to a range of serious complications, such as organ damage, stroke, and heart disease. The International Diabetes Federation (IDF) reports that 10.5% of the adult population aged 20 to 79 are diagnosed with diabetes, and almost half are unaware of the condition. Hence, the number of people with diabetes has increased by fourfold compared to the prior period. One essential step for preventing complications in patients with diabetes is early detection, one of which is by utilizing artificial intelligence (AI) technology, namely data mining. Therefore, knowledge about effective algorithms used to detect diabetes is needed. This study aimed to compare two algorithms, namely k-nearest neighbor (KNN) and support vector machine (SVM), for diabetes classification using the synthetic minority oversampling technique (SMOTE). In this study, both algorithm performance was measured using the machine learning life cycle method. The results showed they had good performance in detecting diabetes; yet, there were significant performance differences between the two. The SVM algorithm with radial basis function (RBF) kernel achieved 81.67% accuracy, 85.91% precision, 79.01% recall, and 82.32% F1 score. Meanwhile, the KNN algorithm with k = 3 found through cross-validation achieved 83.33% accuracy, 85.00% precision, 83.95% recall, and 84.47% F1 score. Based on confusion matrix evaluation, KNN showed superior performance compared to SVM in terms of accuracy and other evaluation metrics. These results indicate that KNN is more effective in detecting diabetes in the dataset used in this study.

Optical Flow Performance in the SUAV Flight Speed Estimation Using Farneback Method

Aziz Fathurrahman, Ony Arifianto, Yazdi Ibrahim Jenie, Hari Muhammad — Thu, 27 Feb 2025 13:08:23 +0700

This paper evaluates the performance of the Farneback optical flow method for estimating the flight speed of a small unmanned aerial vehicle (SUAV) in a simulated 3D World MATLAB-Unreal Engine environment. Optical flow offers a promising solution for velocity estimation, which is crucial for autonomous navigation. A downward-facing monocular camera model was simulated on an SUAV during steady state, straight flight at 100 m altitude and 25 m/s airspeed. Three simulated flight scenes—forest, city block, and water—representing poor, moderate, and rich textures were used to assess the method’s performance. Results demonstrated that using the median estimate of the optical flow field yielded accurate velocity estimations in moderate to rich texture scenes. Over the city block and forest scenes, mean velocity estimation accuracy was 0.6 m/s (σ = 0.2 m/s) and 0.3 m/s (σ = 0.4 m/s), respectively. The impact of camera tilt angle and altitude variations on estimation accuracy was also investigated. Both factors introduced bias, with accuracy decreasing to 1.7 m/s (σ = 0.2 m/s) and 1.9 m/s (σ = 0.2 m/s) for +10° and -10° camera tilt, respectively. Similarly, altitude differences of +10m and -10m resulted in reduced accuracy of 1.9 m/s (σ = 0.2 m/s) and 4.3 m/s (σ = 0.1 m/s), respectively. This study demonstrates the potential of the Farneback method for determining flight speed under steady, straight flight conditions with acceptable accuracy.

Comparison of U-NET and ELU-NET for Pancreatic Cancer Medical Image Semantic Segmentation

Algi Fari Ramdhani, Yudi Widhiyasana, Setiadi Rachmat — Wed, 26 Feb 2025 14:02:11 +0700

Medical image analysis for semantic segmentation using deep learning technology has been extensively developed. One of the notable architectures is U-NET, which has demonstrated high accuracy in segmentation tasks. Further advancements have led to the development of ELU-NET, which aims to enhance model efficiency. ELU-NET achieves relatively good accuracy; however, further comparative analysis of both models is necessary. The comparison between these models is based on accuracy, storage usage, and processing time in performing semantic segmentation of pancreatic cancer images. The pancreatic cancer images utilized in this study are sourced from the PAIP 2023 Challenge, consisting of hematoxylin and eosin (H&E)-stained images. Experiments were conducted by varying the number of filters and model depth for both architectures. The evaluation was performed using a dataset of 57 pancreatic cancer images. The experimental results indicated that U-NET achieved the highest accuracy at 92.8%, slightly outperforming ELU-NET, which attained 89.7%. However, ELU-NET is significantly more efficient in terms of storage usage (8.1 MB for ELU-NET compared to 93.31 MB for U-NET) and processing time (4.0 s for ELU-NET and 5.3 s for U-NET). Although ELU-NET exhibited slightly lower accuracy than U-NET, it surpassed U-NET considerably in terms of storage efficiency (by 85.21 MB) and processing speed (by 1.3 s). These findings suggest that ELU-NET is not superior to U-NET in accuracy. However, given the storage size ratio of 1:11.51 and the processing time ratio of 1:1.325 between ELU-NET and U-NET, the 3.1% accuracy difference represents a reasonable trade-off.

A Multilevel and Hierarchical Approach for Multilabel Classification Model in SDGs Research

Berliana Sugiarti Putri, Lya Hulliyyatus Suadaa, Efri Diah Utami — Wed, 26 Feb 2025 14:09:23 +0700

The growing number of research publications complicates the identification of the implementation of research publications, especially related to sustainable development goals (SDGs). The research publication categorization into SDG levels has not been conducted. The Center for Research and Community Service (Pusat Penelitian dan Pengabdian Masyarakat, PPPM) Politeknik Statistika (Polstat) STIS needs this to monitor lecturers in implementing SDGs. This study aimed to implement and evaluate problem transformation methods and machine learning classification algorithms with a multilevel and hierarchical approach to categorize research publications into SDG levels. Problem transformation methods used were binary relevance, label powerset (LP), and classifier chains. Machine learning classification algorithms used were logistic regression (LR) and support vector machine (SVM). The inputs included titles, abstracts, and titles and abstracts. The best filter model that classified data into SDGs-non-SDGs was the model with titles and SVM, with an accuracy of 0.8634. The best level model for classifying data to SDG level was the model using titles, LP, and SVM with multilevel approaches. The level model classified data into four pillars, goals, targets, and indicators of SDGs, with an accuracy of 0.8067, 0.7501, 0.6792, and 0.6194, respectively. In comparison to other inputs with more comprehensive information, the results showed that title inputs yielded the best accuracy due to the simultaneous use of English and Indonesian. Future research can modify the model to utilize a single language input to optimize the term frequency-inverse document frequency (TF-IDF) process, hence, the word meanings from each language are not considered different important words.

Automatic Liquid-Filling Machine Using Arduino and LabVIEW

Syafriyadi Nor, Zaiyan Ahyadi — Wed, 26 Feb 2025 14:11:37 +0700

The automatic liquid-filling machine plays a vital role in improving efficiency and productivity in modern manufacturing and packaging industries. However, challenges such as high costs, complexity, and limited technical knowledge often hinder its adoption. This research aimed to develop an educational system that is simple, affordable, and easy to implement, helping students grasp the fundamental principles and real-world applications of automatic liquid-filling machines. The system integrates LabVIEW for visual processing and an Arduino Nano microcontroller with the Modbus remote terminal unit (RTU) protocol to simulate industrial communication standards. LabVIEW controls the conveyor belt, filling, and capping processes using ladder logic while recording the number of filled bottles. The Arduino microcontroller manages conveyor belt operations and allows users to set volume and bottle count via a keypad. Serial communication between LabVIEW and Arduino through Modbus RTU provides hands-on experience in configuring industrial systems. Experimental tests under various operational scenarios confirmed the system’s accuracy in filling bottles within a volume range of 250–1,000 ml at a speed of 10 ml/s, handling up to five bottles per cycle. The system demonstrated stable operation without disruptions. This research enhances instrumentation and control system education by offering an interactive, cost-efficient learning tool. The successful use of Modbus RTU underscores its reliability in supporting automatic liquid-filling machines while enriching students’ understanding of industrial automation.

Analysis of Facial Areas to Identify CHD Risks Based on Facial Textures

Budi Sunarko, Agung Adi Firdaus, Yudha Andriano Rismawan, Anan Nugroho — Thu, 27 Feb 2025 13:25:49 +0700

Early screening for coronary heart disease (CHD) remains insufficiently addressed, underscoring the need for a more effective screening tool. Previous studies have reported a classification accuracy of only 72.73%, which is inadequate. This study aimed to develop and evaluate a machine learning model or diagnose CHD using facial texture features and to compare the performance across different facial regions to provide recommendations for improvement. The research involved constructing a machine learning model that extracted texture features from six facial regions of interest (ROIs) using the gray level co-occurrence matrix (GLCM) and employed an artificial neural network (ANN) algorithm. The datasets were full-face images of CHD patients (positive) and healthy people (negative). The face parts identified were the right crow’s feet, right canthus, nose bridge, forehead, left canthus, and left crow’s feet. A total of 132 (72 positive and 60 negative CHD) datasets were divided into 80% (n = 106) training data and 20% (n = 26) testing data. The developed model achieved a notable accuracy of 76.9%. The findings revealed that two facial regions—canthus and forehead—demonstrated excellent accuracy of 80.97% and 90%, respectively. Meanwhile, the crow’s feet and nose bridge regions showed good accuracies at 73.50% and 65%, respectively. Based on the results, this research has proven to be able to become a model for early CHD screening with good accuracy and faster execution.

Perbandingan Penggunaan Optimizer dalam Klasifikasi Sel Darah Putih Menggunakan Convolutional Neural Network

Dede Kurniadi, Rifky Muhammad Shidiq, Asri Mulyani — Thu, 06 Mar 2025 14:11:09 +0700

Sel darah putih adalah komponen penting dari sistem kekebalan tubuh yang bertugas melawan infeksi dan penyakit. Klasifikasi dan penghitungan sel darah putih biasanya dilakukan secara manual oleh operator berpengalaman atau menggunakan analisis sel darah otomatis. Metode manual tidak efisien, memakan waktu, dan membutuhkan banyak tenaga kerja, sedangkan mesin analisis otomatis sering kali mahal dan memiliki persyaratan tinggi untuk sampel uji. Penelitian ini bertujuan untuk membandingkan kinerja tiga jenis optimizer, yaitu RMSProp, SGD, dan Adam, dalam model klasifikasi sel darah putih menggunakan algoritma convolutional neural network (CNN). Dataset yang digunakan terdiri atas 12.392 citra yang mencakup empat kelas sel darah putih, yaitu eosinofil, neutrofil, limfosit, dan monosit. Hasil penelitian menunjukkan bahwa optimizer Adam memberikan kinerja terbaik dengan akurasi pelatihan mencapai 98,65% dan akurasi evaluasi sebesar 97,73%. Adam juga unggul dalam metrik lainnya, seperti recall (97,43%), presisi (97,42%), F1-score (97,42%), dan spesifisitas (99,11%). Nilai AUC untuk setiap kelas melebihi 90%, menunjukkan kemampuan model yang sangat baik dalam membedakan data antarkelas. Optimizer RMSProp menghasilkan akurasi pelatihan sebesar 98,63%, sedangkan SGD menunjukkan akurasi pelatihan lebih rendah, yaitu 83,46%. Penelitian ini menunjukkan bahwa pemilihan optimizer yang tepat berpengaruh pada kinerja CNN dalam klasifikasi citra sel darah putih, sehingga dapat menjadi langkah awal untuk pengembangan sistem klasifikasi medis yang lebih akurat.