Jurnal Transformatika
Not a member yet
319 research outputs found
Sort by
Benchmarking IndoBERT and Transformer Models for Sentiment Classification on Indonesian E-Government Service Reviews
Full text linkThe rapid adoption of e-government services in Indonesia has increased the importance of understanding public sentiment toward digital platforms. This study presents a comparative analysis of five models—IndoBERT, mBERT, XLM-R, CNN, and BiLSTM—for sentiment classification on user reviews of NEWSAKPOLE, a public service application for vehicle tax and licensing. A custom dataset of 11,000+ reviews was scraped from the Google Play Store and labeled using a hybrid rating-based and manual validation approach. Each model was evaluated using accuracy, precision, recall, and F1-score. IndoBERT achieved the highest performance with an F1-score of 0.882, outperforming multilingual and classical deep learning models. Confusion matrix analysis showed that transformer-based models were more effective in detecting neutral and mixed sentiments, while CNN and BiLSTM struggled with misclassification. The results highlight IndoBERT\u27s robustness in low-resource sentiment analysis and its potential to enhance public service monitoring and policy feedback mechanisms in Indonesian digital governance
Algoritma Random Forest, Decision Tree dan XGboost Untuk Klasifikasi Stunting Pada Balita
Full text linkAt the age of toddlers, children need special attention because their brains develop around 80%. Stunting is a form of long-term nutritional deficiency that occurs during the growth and development of children, which are marked with height that is not appropriate or less compared to children their age based on the standard WHO. This condition can adversely affect the cognitive development and health of children. Identifying toddlers who are at risk of experiencing stunting at an early stage is very important to reduce the adverse effects that can affect their quality of life in the future. Traditional methods are less effective in predicting stunting because they often ignore the complex factors that affect the nutritional status of toddlers. This study aims to classify stunting toddlers using Random Forest, Decision Tree, and Extreme Gradient Boost (XGBOOST) algorithms. The results obtained showed that the accuracy of the Random Forest algorithm received the highest accuracy of 99.72 %, Extreme Gradient Boost (XGBOOST) at 99.58 %, and Decision Tree received 98 87 %accuracy.At toddler age, they require special attention, as this is the period when the brain develops up to 80%. Stunting is a form of chronic malnutrition that affects a child\u27s growth and development. According to WHO standards, it is characterised by height that is below or lower than that of peers. This condition has negative impacts on cognitive development and overall health. Identifying toddlers at risk of stunting early is crucial to minimising the negative effects that could impact their quality of life in the future. Traditional methods are less effective in predicting stunting because they often overlook the complex factors that influence an infant\u27s nutritional status. This study aims to compare three algorithms and identify the most effective one for analysing infant stunting data. The method used involves comparing the results of the Random Forest, Decision Tree, and Extreme Gradient Boost (XGBoost) algorithms. The results obtained show that the Random Forest algorithm achieved the highest accuracy at 99.72%, Extreme Gradient Boost (XGBoost) at 99.58%, and Decision Tree at 98.87%
Sistem Pakar Diagnosis Hama Dan Penyakit Tanaman Jeruk Keprok Menggunakan Metode Dempster Shafer Berbasis Web
Full text linkSoe Tangerines are a type of local orange from the Soe area and are one of the leading agricultural commodities in East Nusa Tenggara (NTT) Province. The fruit is round and small with an average diameter of between 5-7 cm. Oranges are an important need for Indonesian people. Although the harvest of these oranges has not yet reached its maximum potential, this is caused by several factors such as cultivation techniques, environmental conditions, as well as pest and disease attacks. These three factors can cause a decrease in productivity and even crop failure. Therefore, an expert system was designed using the Dempster Shafer method. This method aims to detect pests and diseases based on the symptoms that appear on Tangerine plants and provide solutions to overcome these problems. The goal is to reduce the risk of pest and disease attacks on plants. The test results used 50 case data from the TTS Regency Agriculture and Plantation Service. The system test results from 50 data produced 42 case data that were in accordance with the expert diagnosis, 6 data that were in accordance with the expert diagnosis but below the threshold of 80% and 8 data that were not appropriate. Testing system accuracy by comparing system and expert diagnosis results obtained accuracy results of 84%
Perbandingan Naïve Bayes dan K-NN dalam Analisis Sentimen Aplikasi X
Full text linkAplikasi X, sebelumnya dikenal sebagai Twitter adalah media sosial yang memungkinkan pengguna mengirim, membalas, dan membaca pesan. Berdasarkan ulasan di Google Play Store, banyak pengguna mengeluhkan masalah, terutama terkait penangguhan akun setelah perubahan kepemilikan. Namun, sebagian pengguna masih merasa puas dan terbantu dengan X. Oleh karena itu, analisis sentimen dilakukan untuk mengetahui kecenderungan opini pengguna. Penelitian ini menggunakan metode naïve bayes dan k-Nearest Neighbor pada 8.723 ulasan yang kemudian diklasifikasi sebagai sentimen positif, netral, atau negatif menggunakan K-fold cross validation. Naïve Bayes mencapai akurasi tertinggi sebesar 88,87% pada 10-fold, sementara KNN dengan k optimal di 12-NN mencapai 90,32% pada 2-fold. Dalam perbandingan hasil klasifikasi dengan label pakar kedua, metode Naïve Bayes lebih sesuai dengan akurasi 92,56% dibandingkan KNN yang mencapai 91,73%.Aplikasi X, sebelumnya dikenal sebagai Twitter adalah media sosial yang memungkinkan pengguna mengirim, membalas, dan membaca pesan. Berdasarkan ulasan di Google Play Store, banyak pengguna mengeluhkan masalah, terutama terkait penangguhan akun setelah perubahan kepemilikan. Namun, sebagian pengguna masih merasa puas dan terbantu dengan X. Oleh karena itu, analisis sentimen dilakukan untuk mengetahui kecenderungan opini pengguna. Penelitian ini menggunakan metode naïve bayes dan k-Nearest Neighbor pada 8.723 ulasan yang kemudian diklasifikasi sebagai sentimen positif, netral, atau negatif menggunakan K-fold cross validation. Naïve Bayes mencapai akurasi tertinggi sebesar 88,87% pada 10-fold, sementara KNN dengan k optimal di 12-NN mencapai 90,32% pada 2-fold. Dalam perbandingan hasil klasifikasi dengan label pakar kedua, metode Naïve Bayes lebih sesuai dengan akurasi 92,56% dibandingkan KNN yang mencapai 91,73%
Design and Development Optimized FIFO Queue System for Food Outlets
Full text linkIn today\u27s fast-paced food service industry, the efficiency of queue management is vital to operational success, profitability, and customer satisfaction. This study evaluates an integrated queue management system\u27s impact on these critical areas. The results show an average satisfaction score of 80.11% from customers and 90.37% from food outlet owners, demonstrating the system\u27s strong effectiveness. The research focused on the importance of reducing perceived waiting times through real-time updates, which enhance customer tolerance and satisfaction. By combining online and onsite ordering, the system provides real-time updates, order tracking, and notifications to boost efficiency and minimize cancellations. Despite some identified weaknesses, such as the absence of direct customer reviews and existing bugs, the system holds significant potential for improving user experience. These findings highlight the necessity for continuous development and maintenance to optimize the system further. Overall, this approach promises to advance the operational capabilities and customer satisfaction levels of food outlets
PENGARUH PENERAPAN ROUTING I-BGP TERHADAP WAKTU FAILOVER DALAM JARINGAN LOKAL
Full text linkKeandalan jaringan komputer menjadi elemen penting dalam mendukung berbagai sektor di era digital. Failover, mekanisme pengalihan otomatis ke jalur cadangan saat jalur utama gagal, sangat bergantung pada kecepatan respons dan kemampuan protokol routing. Penelitian ini mengevaluasi efektivitas Internal Border Gateway Protocol (I-BGP) dalam mempercepat waktu failover pada jaringan lokal berbasis Mikrotik yang terhubung melalui VPN. Performa I-BGP dibandingkan dengan OSPF, RIP, dan Static Routing melalui pengujian waktu failover, jumlah paket hilang, dan efisiensi bandwidth. Hasil menunjukkan I-BGP memiliki waktu failover tercepat (0,51 detik), kehilangan paket minimal (2 paket), dan utilisasi bandwidth tertinggi (95%). Uji ANOVA mengonfirmasi perbedaan signifikan antar protokol (F=776,898, p<0,001). Temuan ini menegaskan keunggulan I-BGP sebagai solusi optimal untuk failover cepat dan andal. Dimasa mendatang tantangan dan kompleksitas dari jaringan internet akan semakin besar, untuk itu perlu diadakan kajian komprehensif dan penelitian lebih lanjut tentang gangguan konektifitas yang berkaitan dengan faktor keamanan, dan solusi dengan penggunaan kecerdasan buatan.The reliability of computer networks is a crucial element in supporting various sectors in the digital era. Failover, an automatic switching mechanism to a backup route when the main route fails, heavily depends on response speed and the capability of routing protocols. This study evaluates the effectiveness of Internal Border Gateway Protocol (I-BGP) in accelerating failover time on a local network based on Mikrotik connected via VPN. The performance of I-BGP is compared with OSPF, RIP, and Static Routing through tests on failover time, packet loss, and bandwidth efficiency. The results show that I-BGP has the fastest failover time (0.51 seconds), minimal packet loss (2 packets), and the highest bandwidth utilization (95%). ANOVA testing confirms significant differences among the protocols (F=776.898, p<0.001). These findings highlight the superiority of I-BGP as an optimal solution for fast and reliable failover. In the future, as network complexity and challenges increase, comprehensive studies and further research on connectivity disruptions related to security factors and AI-based solutions will be necessary
Komparasi AHP, SAW, TOPSIS, VIKOR, dan MABAC pada Sistem Pengambilan Keputusan Pemilihan Supplier Obat
Full text linkThe selection of pharmaceutical suppliers is crucial for ensuring consistent drug availability and maintaining service quality in healthcare facilities. This study offers a comparative analysis of five Multi Criteria Decision Making methods (AHP, SAW, TOPSIS, VIKOR, and MABAC) applied to supplier evaluation based on four key criteria: price, delivery time, receipt accuracy, and product quality. Unlike previous studies that employed individual or dual methods, this research evaluates all five methods using the same dataset to assess consistency, sensitivity, and decision reliability. The results show strong ranking consistency across methods, with AHP and SAW producing identical outputs. TOPSIS and VIKOR offer similar outcomes based on proximity and compromise analysis, while MABAC demonstrates high discrimination power for mid-ranked suppliers. Sensitivity tests confirm ranking stability under moderate weight variations. This study provides practical recommendations for selecting appropriate decision methods in pharmaceutical procurement systems based on operational context and desired decision accuracy.Pemilihan supplier obat yang tepat merupakan aspek strategis dalam manajemen apotek. Penelitian ini bertujuan membandingkan kinerja 5 metode Multi-Criteria Decision Making (MCDM), yaitu AHP, SAW, TOPSIS, VIKOR, dan MABAC, dalam sistem pendukung keputusan untuk pemilihan supplier obat. Studi ini melibatkan data 10 supplier berdasarkan 4 kriteria utama: harga, pengiriman, penerimaan barang, dan kualitas. Hasil menunjukkan bahwa metode AHP dan SAW memberikan hasil identik dan stabil, sedangkan TOPSIS dan VIKOR unggul dalam mengakomodasi kompromi antar kriteria. MABAC menunjukkan sensitivitas tinggi dalam membedakan kinerja supplier di peringkat menengah ke bawah. Hasil Uji sensitivitas perubahan bobot kriteria harga dengan kenaikan 10% dan 20% atau penurunan 10% dan 20% dapat berpengaruh pada pemeringkatan supplier. Penelitian ini merekomendasikan pemilihan metode berdasarkan kebutuhan sistem, di mana AHP dan SAW cocok untuk implementasi manual. Temuan ini diharapkan dapat membantu apotek dalam mengambil keputusan lebih objektif dan efisien
Analisis Pembelajaran Statistika Berbasis Literasi menggunakan E-Modul Berbantuan Flipbook
Full text linkThis research investigates the implementation of literacy-based learning in Statistics through the development of an e-module enhanced with Flipbook technology, aiming to foster student comprehension and active participation. The study employs a Research and Development (R&D) approach, utilizing the 4D model consisting of the phases: Define, Design, Develop, and Disseminate. The main output is a literacy-focused e-module presented in Flipbook format, specifically designed for students in the Informatics Engineering department. Validation from subject matter and media experts confirmed the module’s quality, rating it as “feasible” to “highly feasible.” Feedback from students reflected strong approval, with an overall response score averaging 83.75%, falling into the “very good” category. The use of the module significantly enhanced students’ grasp of statistical concepts, as evident from the improved scores in both pretest and posttest assessments. The recorded average gain score of 0.70, classified as “high,” demonstrates the e-module’s effectiveness in boosting learning outcomes. This integration of technology and literacy within the e-module makes it an appropriate and impactful learning tool to enrich the Statistics learning experience in a more engaging and contextually relevant manner
Komparasi Algoritma Random Forest dan XGBoost dalam Prediksi Premi Asuransi Kesehatan
Full text linkAsuransi kesehatan saat ini menjadi salah satu hal yang banyak orang persiapkan dikarenakan adanya ketidakpastian risiko kesehatan dan biaya layanan kesehatan yang semakin naik. Perhitungan premi tiap individu dapat berbeda dikarenakan terdapat perbedaan profil kesehatan seperti usia, BMI maupun gaya hidup seperti merokok yang membuat perusahaan asuransi harus memperhitungkan premi dengan akurat agar tidak menimbulkan kerugian finansial dan sesuai dengan tingkat risiko terjadinya klaim. Adapaun tujuan dari penelitian ini adalah melakukan komparasi antara algoritma Random Forest dan XGBoost dalam memprediksi premi asuransi kesehatan berdasarkan beberapa faktor yang sulit dihitung secara manual. Evaluasi dilihat berdasarkan metrik regresi yaitu MAE, MSE, RMSE, dan R2. Pada penelitian ini, algoritma Random Forest berhasil memprediksi premi asuransi kesehatan lebih baik dari XGBoost dengan nilai MAE 2.573, MSE 24199792,43 RMSE 4919, 33 dan R2 sebesar 84.04%
Penggunaan Feature Space SMOTE Untuk Mengurangi Overfitting Akibat Imbalance Dataset
Full text linkThe creation of a classification model requires careful consideration of several crucial factors to achieve optimal performance. A good model is typically indicated by high accuracy and F1-score values, as well as low loss values. To create a successful model, certain conditions must be met, including selecting the appropriate architecture and ensuring the availability of high-quality data. In this study, a classification model for CT Kidney Stone was developed using an imbalanced dataset obtained from Kaggle. The chosen algorithm for model development was Convolutional Neural Network (CNN), as CNN is known for its effectiveness in image classification tasks. Three different pre-processing approaches were employed in model creation. The first model was built using the imbalanced training data. The second model involved data augmentation, while the third model utilized SMOTE oversampling. Subsequently, all three models were evaluated using private data to assess testing performance and identify any potential overfitting. The research findings revealed that the third model exhibited the best performance among the three, showcasing its superiority in handling the imbalanced dataset and achieving optimal results.The creation of a classification model requires careful consideration of several crucial factors to achieve optimal performance. A good model is typically indicated by high accuracy and F1-score values, as well as low loss values. To create a successful model, certain conditions must be met, including selecting the appropriate architecture and ensuring the availability of high-quality data. In this study, a classification model for CT Kidney Stone was developed using an imbalanced dataset obtained from Kaggle. The chosen algorithm for model development was Convolutional Neural Network (CNN), as CNN is known for its effectiveness in image classification tasks. Three different pre-processing approaches were employed in model creation. The first model was built using the imbalanced training data. The second model involved data augmentation, while the third model utilized SMOTE oversampling. Subsequently, all three models were evaluated using private data to assess testing performance and identify any potential overfitting. The research findings revealed that the third model exhibited the best performance among the three, showcasing its superiority in handling the imbalanced dataset and achieving optimal results