Jurnal CoSciTech (Computer Science and Information Technology)

Implementation of an IoT-Based Smart Cane Using YOLO V3 to Enhance the Mobility and Safety of Visually Impaired Individuals

Dava Febrian Dava — 2025-08-04

Visually impaired individuals face significant challenges in mobility and safety during daily activities, especially in public spaces that are not disability-friendly. Conventional white canes are limited in their ability to detect obstacles. This study aims to design and implement a smart cane based on the Internet of Things (IoT) and the real-time object detection algorithm YOLO V3 to enhance the mobility and safety of visually impaired users. The developed system utilizes ultrasonic sensors to detect obstacles on the left and right sides of the user, a GPS module for real-time location tracking via a web server, and an ESP32-CAM integrated with YOLO V3 to detect objects such as vehicles, holes, and people. Information is conveyed to the user through voice alerts using a DFPlayer Mini and is also displayed on an LCD and a web interface. Test results show that the system operates accurately, with an average sensor error rate of only 0.12%, and all components function properly. Usability testing involving 50 respondents indicates a very high level of user satisfaction, with average agreement rates exceeding 85%. This research demonstrates that the integration of IoT and computer vision can produce a smart, responsive, and user-friendly assistive device for the visually impaired.

Implementation of the XGBoost Algorithm for Predicting Monthly Regional Revenue Performance in Bandung City

Puteri Marchanda Izzati — 2025-08-04

Abstract

Local Own-Source Revenue (PAD) is a key pillar in financing regional development. In Bandung City, discrepancies between revenue targets and actual realization remain a challenge to effective fiscal planning. This study aims to develop a predictive model for monthly PAD achievement using the XGBoost algorithm, known for its strength in handling non-linear and complex data. The dataset, obtained from the Bandung Revenue Agency (Bapenda), includes various types of regional taxes from 2018 to 2024. The research process involved data cleaning, feature engineering, data splitting, model training, and performance evaluation using MAE, RMSE, and R² metrics. The evaluation on test data resulted in MAE of IDR 5.6 billion, RMSE of IDR 9.3 billion, and R² of 73%. Meanwhile, 5-fold cross-validation yielded MAE of IDR 3.49 billion, RMSE of IDR 6.65 billion, and R² of 86%. These results demonstrate high accuracy and generalization capability. XGBoost proves to be a reliable decision-support tool for data-driven fiscal planning.

Rancang Bangun Sistem Informasi Pimpinan Daerah 114 Tapak Suci Putera Muhammadiyah Pekanbaru Menggunakan Model Waterfall

Muhammad Ryan Pratama Yudha — 2025-08-04

In Tapak Suci, particularly in the Pekanbaru region, the organizational leadership structure in Indonesia comprises four levels: the school (perguruan), branch leadership (pimpinan cabang), regional leadership (pimpinan daerah), provincial leadership (pimpinan wilayah), and central leadership (pimpinan pusat). Currently, the management of member data, ranking data, and other related information is still conducted manually. This process is often time-consuming, inaccurate, and difficult to retrieve. Consequently, the organization's management faces challenges in obtaining fast and accurate information to support decision-making. Furthermore, manual data management is prone to human error and data loss.Information system technology offers a potential solution to these challenges. This research uses a Research and Development (R&D) methodology by using an integrated information system to enable more efficient, accurate, and structured data and information management. The development process adheres to the Waterfall paradigm, which comprises the phases of system design and requirement analysis., coding, testing, and system maintenance.The result of this research is a web-based information system for managing the activities of 114 Regional Leaderships (Pimpinan Daerah) of Putera Muhammadiyah. This data management system is designed for two types of users: PIMDA administrators and Branch administrators. The information system is expected to improve the efficiency of organizational management.

Classification of Avocado Plant Varieties Based on Leaf Shape Using CNN Algorithm

Agum Pratama — 2025-08-04

The avocado plants is a popular horticultural commodities in Indonesia, especially in Java, due to their health benefits and high economic value. However, differences in leaf shape across avocado varieties often make identification difficult for both buyers and sellers, which can lead to transaction errors and losses. Manual identification requires specialised skills that are not always available, especially in areas such as Kuningan Regency. To answer these problems, this research aims to develop an Android-based application that is able to classify avocado varieties, namely alligator, kendil, and butter, based on leaf images automatically. This application uses Convolutional Neural Network (CNN) algorithm with SSDMobileNetV2 FPNLite pre-trained model implemented through TensorFlow framework. The dataset used consists of 4,800 avocado leaf images divided for training, validation, and testing processes. The test results show that the model is able to achieve an accuracy rate of 99%. For the alligator class, the precision and recall values were 1.00 and 0.98 respectively; for the kendil class, 1.00 and 0.99; and for the butter class, 0.99 and 1.00. These findings prove that the CNN algorithm is effective in classifying avocado varieties based on visual characteristics of the leaves. Thus, this application has the potential to become a fast, accurate, and practical tool in the process of identifying avocado varieties, both for commercial and educational purposes.

Design and Construction of Reseller Information System with Web-Based Customer Relationship Management Approach

Ruri Mutiara Ayuni — 2025-08-04

This study aims to design and implement a web-based reseller information system at Mutiara Cell with a Customer Relationship Management (CRM) approach to answer the needs of digital transformation in improving operational efficiency and managing business relationships. Mutiara Cell is a company that distributes electric credit and digital products that involves many reseller partners, but still faces obstacles in recording transactions, managing data, and reporting finances that are done manually. The system developed is designed to facilitate administrators and resellers in ordering products, uploading proof of payment, managing bills, and reporting sales digitally and in an integrated manner. The development uses a waterfall model with stages of needs analysis, design, implementation, and testing, and uses PHP Native as a programming language and Black Box Testing to ensure system functionality. The results show an increase in operational efficiency, data access speed, and accuracy in transaction management. In addition, the user-friendly interface supports convenience in accessing services, and the CRM approach has been proven to strengthen business relations and increase partner loyalty. This system is considered feasible to use and is a long-term digital solution in supporting operational activities and managing relationships with resellers more professionally.

Classification of philodendron plant species based on leaf images using a CNN algorithm

Muhammad Alif Fathan Alif — 2025-08-06

Accurate identification of ornamental plants is becoming important as public interest in tropical plant collections increases, one of which is from the Philodendron genus. This ornamental plant has many varieties that are often difficult to distinguish due to visual similarities in the shape and pattern of their leaves. This research aims to develop a system for Philodendron type classification based on leaf images using the Convolutional Neural Network (CNN) algorithm to help the identification process. The method used is with a dataset of 5000 leaf images of five Philodendron species, which are divided into 80% training data, 10% validation data, and 10% test data. A CNN model with MobileNetV2 FPNLite SSD architecture was implemented and trained for 50,000 steps, then optimised for mobile devices using TensorFlow Lite. Performance analysis was conducted using confusion matrix to evaluate accuracy, precision, recall, and F1-Score metrics. The results show that the developed model is able to accurately classify leaf images, both in the form of static images and in real-time. This system has been successfully implemented in an Android application that is expected to be a practical identification tool for general users and ornamental plant enthusiasts.

Application of Learning Vector Quantization in Digital Image Processing for Skin Disease Detection

Rizki Akbar Pratama — 2025-08-10

Skin, as the largest human organ, covers more than two square meters and accounts for about 15% of body mass. Consisting of three main layers of epidermis, dermis, and subcutaneous tissue, the skin serves as a physical shield and barrier against infection, injury, and UV radiation. Skin diseases such as chickenpox, monkey pox, measles and herpes are medical challenges that require quick and accurate diagnosis. This study used 520 digital images (130 per category) from Mendeley Data and online sources. The Learning Vector Quantization (LVQ) algorithm was applied for image classification based on the extracted features. Results showed an overall accuracy of 90.74%, with respective accuracies: 97% (chickenpox), 98% (monkey pox), 91% (measles), and 100% (herpes). Evaluation using confusion matrix resulted in accuracy, precision, recall, and F1-score values of 0.91, indicating strong model performance. These findings demonstrate the potential of LVQ as a digital image-based skin disease diagnosis tool.

Predictive Modeling of Diabetes Using Multimodel Machine learning and Deep learning Approaches

Fadli Rahmad Hidayatullah — 2025-08-10

This study discusses the implementation and evaluation of various machine learning algorithms along with one deep learning model for predicting diabetes based on patient medical data. The dataset underwent Preprocessing steps including categorical feature Encoding, feature scaling, and train-test split. The algorithms compared in this study include Logistic regression, Decision Tree, Random Forest, and K-Nearest Neighbors (KNN). Additionally, a Multilayer Perceptron (MLP) model was developed using Keras to explore a deep learning approach with the use of epochs and batch size. The model performance was evaluated using accuracy, precision, and recall metrics, along with learning curve visualizations to analyze model convergence during training. The evaluation results showed that the Random Forest model achieved the highest accuracy among traditional algorithms, while the MLP provided competitive results with strengths in generalization. Visualization of loss and accuracy per epoch offered deeper insight into model behavior throughout the training process. This study demonstrates that a combination of proper data Preprocessing techniques and appropriate model selection significantly influences prediction accuracy. The findings may serve as an early reference for the development of data-driven medical prediction systems and support computer-assisted clinical decision-making (clinical decision support systems).

Identifikasi penyakit tanaman tomat melalui citra daun menggunakan DenseNet201

Okamisar — 2025-08-13

This study focuses on implementing the DenseNet201 algorithm for disease classification in tomato plants using leaf images from PlantVillage dataset. The agricultural sector plays a central role in the Indonesian economy, with tomatoes being one of the important horticultural crops. However, tomato productivity is often hindered by various plant diseases. Accurate disease diagnosis is crucial for improving production stability. Image processing-based approaches, such as Convolutional Neural Network (CNN), have facilitated effective plant disease diagnosis. In this study, the PlantVillage dataset consisting of 18,835 tomato leaf images is utilized. The data is divided into training (10,000 images), validation (7,000 images), and test (500 images) sets. A classification model is constructed using the DenseNet201 architecture with some modifications. The results show that the DenseNet201 model achieves an accuracy of 95.20% on the testing data, with an overall F1-score of 0.95. Compared to previous studies using VGG16 (77.2%), InceptionV3 (63.4%), and MobileNet (63.75%), the DenseNet201 model demonstrates a significant performance improvement. This study concludes that DenseNet201 is highly effective in classifying tomato plant diseases and has the potential to be implemented in widespread plant disease diagnosis applications.

Implentation of CNN for Corn Leaf Disease Identification

Gilang Gumelar — 2025-08-24

Maize is an important commodity in Indonesia's agricultural sector. However, disease attacks on the leaves can reduce the quality and quantity of the harvest. At SMK Negeri 1 Kuningan, disease identification is still done manually, so there is a risk of errors. This research aims to design and build an Android application to automatically detect corn leaf diseases using the Convolutional Neural Network (CNN) algorithm. The development method used is Rapid Application Development (RAD), with a CNN model based on MobileNetV2 architecture trained using a dataset of diseased and healthy corn leaf images. Evaluation using test images resulted in an accuracy of 96.2%. The model was able to detect five categories: leaf spot, downy mildew, leaf blight, leaf rust, and healthy leaves. The F1-Score is 94% Leaf Spot, 96% Leaf Blight, 96% Healthy Leaf, 97% Leaf Blight, and 96% Leaf Rust, respectively. The precision and recall values of all classes are above 94%. These results show that the integration of CNN in mobile applications is effective in helping the automatic identification of corn leaf diseases in an educational environment.

Correlation between daily COVID-19 cases and vaccine company stock movements in the global market using Long Short-Term Memory (LSTM)

Gigih Setyaji — 2025-08-24

The COVID-19 outbreak has had a significant impact on stock price fluctuations in the pharmaceutical industry, particularly among vaccine-producing companies. This study evaluates the relationship between the number of daily COVID-19 cases and the stock price movements of global vaccine companies, with a primary focus on AstraZeneca (AZN). The predictive model employed is Long Short-Term Memory (LSTM), a deep learning algorithm based on time series data. To achieve more accurate predictions, automatic hyperparameter tuning was performed using the Optuna method. Based on the evaluation results, the model demonstrated high predictive performance, with a Mean Squared Error (MSE) of 1.131777, Mean Absolute Error (MAE) of 0.773518, Root Mean Squared Error (RMSE) of 1.063850, and a coefficient of determination (R²) of 0.974614. Additionally, the model was able to realistically forecast the AZN stock price trend for the next 30 days. These results prove that the optimized LSTM model can serve as an effective prediction tool for analyzing the impact of the pandemic on the capital market.

Implementasi Metode Prototype dalam Pengembangan Sistem Informasi Inventaris Obat di Apotek Syira Farma.

Vicky Setia Gunawan — 2025-08-25

Apotek Syira Farma merupakan sebuah entitas bisnis yang berdedikasi dalam penjualan serta penyediaan berbagai jenis obat-obatan untuk memenuhi kebutuhan kesehatan masyarakat. Apotek ini berperan penting sebagai salah satu fasilitas kesehatan primer yang menyediakan akses mudah terhadap produk farmasi yang aman dan berkualitas. Saat ini, Apotek Syira Farma masih mengandalkan sistem persediaan manual. Artinya, setiap hari para pegawai mencatat seluruh transaksi penjualan dan setiap pemasukan stok barang dari pemasok langsung ke dalam buku laporan persediaan. Kemudian, pimpinan apotek akan memeriksa laporan manual ini secara rutin setiap harinya untuk memantau pergerakan stok. Namun, sistem manual ini jauh dari kata efektif dan efisien. Pengelolaan persediaan secara manual sangat rentan terhadap berbagai kesalahan. Seringkali, terjadi ketidaksesuaian antara data yang tercatat dengan kondisi fisik barang, baik itu kesalahan pencatatan barang masuk, barang keluar, hingga barang kedaluwarsa yang luput dari pendataan. Akibatnya, pimpinan apotek sering kesulitan saat mencoba mencocokkan data persediaan di buku dengan jumlah barang riil yang ada di rak, sehingga proses pengambilan keputusan pun jadi terhambat. Oleh karena itu, sistem inventaris dirancang untuk mengatasi permasalahan tersebut, dengan harapan dapat membantu Apotek Syira Farma mengelola data persediaan secara lebih cepat dan akurat. Penerapan sistem informasi ini memiliki beberapa tujuan utama: meningkatkan kinerja pegawai dalam mengelola stok obat, mempermudah proses rekapitulasi data penjualan, serta memudahkan pimpinan dalam memantau dan menganalisis laporan penjualan dengan lebih tepat dan real-time.

Comparison of Machine Learning Models for Classification and Detection of Heart Disease

Tengku Fawwaz Fatihul Ihsan — 2025-09-01

Heart disease is one of the leading causes of death in the world, so early detection is an important aspect in prevention efforts. This study aims to build a heart disease risk prediction model based on patient clinical data using the Random Forest algorithm. The dataset used consists of 303 data with 13 features such as blood pressure, cholesterol, maximum heart rate, and others, as well as one nested target attribute. The data processing process includes cleaning invalid values such as question marks ('?') which are changed to missing values, and deleting incomplete data to maintain the integrity of the dataset. After going through data exploration and correlation analysis between features, the model is trained using the Random Forest algorithm because of its ability in multiclass classification and resistance to overfitting. The initial evaluation results show that the model has good prediction accuracy with a score reaching 0.89. This study proves that the Random Forest-based machine learning approach is effective in helping the process of systematically identifying heart disease risks, so it has the potential to be a decision support tool in the field of preventive health.

Web Application Security Risk Assessment Using ISO/IEC 27005:20022 Standard for Organizational Services

Nungky Chandra — 2025-09-03

The problem of information security vulnerability and threat risks is increasing, so it is necessary to be able to analyze the risk situation of future information security threats and vulnerabilities, especially for application services of a community organization. Research on the application of information security risk analysis based on the ISO/IEC 27005: 2022 framework in an organization's service applications. ISO/IEC 27005: 2022 is an international standard used for guidelines for implementing the most effective information security risk analysis process compared to other information security risk assessment method frameworks. The results of the assessment are to measure the level of information security risk of an organization's service application so that it can be used as material for improvements in carrying out information security prevention and control measures so that vulnerability gaps and threats of information security attacks can be reduced. The results of this study can describe the risk value in the organization's service application with 3 high-risk categories, namely in financial transaction data (risk value 20), customer database (risk value 16), and server configuration (risk value 15). And medium risk values are found in public APIs (risk value 12) and internal report data (risk value 6).

Implementation of Vision Transformers Model in Website-Based Facial Skin Type Classification

Dila Aura Futri — 2025-09-03

Skin type misidentification often leads to inappropriate skincare product selection, which can negatively affect skin health. This study aims to develop a web-based automatic facial skin type classification system using the Vision Transformer (ViT) architecture. The model implemented is ViT Base Patch 16, pre-trained on the ImageNet dataset and fine-tuned using 10,000 facial images evenly distributed across four classes: normal, dry, oily, and combination. The dataset underwent augmentation and normalization during preprocessing. The training results showed an accuracy of 78% on the test data, with the best performance in the combination skin class (F1-score of 0.86) and the lowest in the normal skin class (F1-score of 0.72). The model was integrated into a Flask-based system that enables users to classify their skin type by either uploading an image or capturing it via camera. System testing was conducted using functional testing and API testing via Postman. The results demonstrated that all key features of the system functioned properly, and the API successfully returned classification responses in JSON format. This system can assist users in identifying their skin type and serve as a reference for selecting appropriate skincare ingredients.

Recommendation Implementation of a Digital Book Recommendation System Using Item-Based Collaborative Filtering in a University Library Application.

Mutsna Mutsna — 2025-09-03

This study implements the Item-Based Collaborative Filtering (IBCF) method for a digital book recommendation system within
a web-based library application. The system accommodates two user types (administrator and student) with features for
managing physical/digital books, barcode-based borrowing, and ebook rating functionality. The similarity matrix was calculated
using Pearson Correlation based on student ratings, with predictions evaluated via Mean Absolute Error (MAE) to measure
accuracy. Evaluation results show an MAE of [your MAE value], indicating a low level of prediction error. Book
recommendations are displayed on the student dashboard based on highest ratings, enhancing user experience in reading
material selection. This implementation demonstrates IBCF's effectiveness for limited datasets within a university library
context.

Decision Support System for Selecting the Best Students in the Faculty of Science and Technology, Muhammadiyah University of East Kalimantan Using the AHP Method

Fachri Boy — 2025-09-04

Penelitian ini bertujuan untuk mengetahui pengaruh pemilihan mahasiswa terbaik pada Fakultas Sains dan Teknologi. Pemilihan mahasiswa terbaik dapat dinilai berdasarkan IPK. Untuk membantu pihak universitas dalam menentukan mahasiswa terbaik, maka dibuatlah Sistem Pendukung Keputusan dengan menggunakan kriteria IPK, TOEFL, LIFE SKILLS, dan MENGAJI dengan menggunakan metode AHP. Metode ini digunakan karena salah satu kemungkinan penyelesaian dari permasalahan tidak terstruktur. Hasil dari penentuan ini adalah mahasiswa dengan predikat tertinggi merupakan mahasiswa dengan lulusan terbaik. Objek penelitian ini adalah Fakultas Sains dan Teknologi di Universitas Muhammadiyah Kalimantan Timur. Jenis penelitian ini adalah penelitian kuantitatif. Populasi dalam penelitian ini adalah nama-nama mahasiswa yang lulus dalam dua tahun terakhir dari tahun 2021-2022 sampai dengan tahun 2022-2023 di Fakultas Sains dan Teknologi. Sampel penelitian dipilih dengan menggunakan Analytic Hierarchy Process. Penelitian ini menggunakan data sekunder yang diperoleh dalam penelitian ini melalui data MKDU dan Program Studi Bahasa Inggris.

Comparison Of Random Forest Regressor And Decision Tree Regressor For Crop Yield Prediction

Rizki Faizal — 2025-09-05

Uncertainty in crop yields due to environmental factors remains a major challenge in Indonesia's agricultural sector. This study aims to compare the performance of the Random Forest Regressor and Decision Tree Regressor algorithms in predicting cultivated crop yields. The dataset used was sourced from Kaggle, consisting of 300,000 rows with features such as crop type, soil type, rainfall, fertilizer use, irrigation, and weather conditions. The system was developed using Python and Streamlit. The methodology includes data preprocessing, model training, and evaluation using the Mean Absolute Error (MAE) metric. The test results show that the Decision Tree Regressor achieved a lower MAE (0.43) compared to the Random Forest Regressor (0.48), resulting in more accurate predictions on this dataset. Feature analysis indicates that rainfall and crop type are the most influential factors. Although Random Forest is generally known for its stability, this study demonstrates that Decision Tree can outperform it within the context of the dataset used. The developed system is expected to assist farmers and policymakers in planning agricultural production more efficiently and in a data-driven manner.

Potato Leaf Disease Classification with Transfer Learning Using CNN Optimization of MobileNetV2 Architecture

Rahmad Gunawan — 2025-09-05

Potatoes are a major food crop with high economic value, but they are susceptible to various Diseases impacting potato leaves can significantly influence their quality and productivity. This research focuses on identifying diseases in potato leaves through the Convolutional Neural Network (CNN) approach, leveraging transfer learning with the MobileNetV2 architecture. The dataset utilized comprises 4,072 images of potato leaves. categorized into three groups: non-infected leaves (healthy ), Early Blight-infected leaves, and Late Blight-infected leaves. The dataset is processed through data augmentation and normalization to enhance data quality. The resulting model demonstrates excellent performance, achieving an accuracy of 95.31%, a precision of 95.81%, a recall of 95.31%, and an F1-Score of 95.38%. These findings indicate the approach demonstrates its ability to identify the condition of potato leaves with a low classification error rate, especially in the healthy category. However, there are challenges in classifying between Early Blight and Late Blight that require further analysis and method improvement. This study contributes to the development of efficient and accurate plant disease detection systems.

Perancangan sistem informasi tiket booking online travel umrah studi kasus: PT.Sukkari Halal Tour

abrar raja — 2025-09-08

This study focus to design an online umrah ticket booking information system of PT.Sukkari Halal Tour to enhance service efficiency and information transparency. The Prototype method was employed, covering stages such as requirement gathering, prototype development, evaluation, coding, testing, and implementation. The result include the design of a use case diagram, activity diagram, sequence diagram, and class diagram, which comprehensively illustrate the system flow. The system accomondates two actor: pilgrims and administrator with features like online registration, document verification, digital payment, and travel packages management. The research concludes that this system simplifies administrative processes and improves user experience.

Optimasi Metode Certainty Factor Menggunakan Rank Order Centroid Pada Sistem Pakar Pendeteksi Turnover Intention Berbasis WEB

Muhammad Maulana Akbar — 2025-09-13

Turnover intention, or the tendency of employees to resign, poses a significant challenge for companies—especially when dealing with Generation Z, who tend to have lower job commitment and are more likely to switch jobs. This study aims to develop a web-based expert system to detect the level of employee turnover intention by integrating the Certainty Factor (CF) and Rank Order Centroid (ROC) methods. The CF method is used to handle uncertainty in questionnaire assessments, while ROC is implemented to optimize the weights among aspects, namely Thinking of Quitting, Intention to Search for Alternatives, and Intention to Quit. The system is built based on 36 questionnaire statements and tested on 34 respondents. The results show that the system provides more proportional and realistic interpretations compared to the non-optimized approach. Accuracy testing indicates that 27 out of 34 system results match manual assessments, yielding an accuracy rate of 79.41%. These findings suggest that the system performs reliably and can serve as a practical tool for the early detection of turnover intention in the workplace.

Rancang Bangun Aplikasi Pengolahan Data Nilai Siswa Berbasis Web dengan Metode Agile pada Sekolah Dasar 011 Kebun Agung

Muhamad Aji Romadhon Aji — 2025-09-13

Elementary schools are formal educational institutions for children aged 6–12 years and play a crucial role in the early stages of education. With the rapid advancement of information technology across various sectors, including education, the integration of technology into school administrative processes has become increasingly important—particularly in managing student grade data. This study was conducted to provide a solution for SD 011 Kebun Agung by developing a web-based student grade management system that enables faster, more accurate, and more accessible input, processing, and dissemination of grade information for teachers and parents. The application was designed and developed using the Agile software development methodology, with Python as the programming language and Django as the framework, and was deployed on the PythonAnywhere platform for online access. Testing using the black-box method indicated that all system features operated effectively in accordance with user requirements. Moreover, interviews with teachers indicated that the application significantly improved the efficiency of grade processing and optimized the communication of academic information to parents. Therefore, the system developed through this study successfully accommodates the needs of SD 011 Kebun Agung in enhancing the efficiency and effectiveness of student grade data management.

Classification of Cucumber Leaf Diseases Based on Convolutional Neural Network (CNN)

Maryogi Yanto — 2025-09-13

Penyakit daun pada tanaman mentimun merupakan salah satu tantangan utama dalam meningkatkan hasil panen, terutama di Kalimantan Barat. Identifikasi penyakit secara manual seringkali tidak akurat dan memakan waktu. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi otomatis untuk penyakit daun mentimun berbasis Convolutional Neural Network (CNN) menggunakan arsitektur VGG-16. Dataset terdiri dari 2.000 citra daun mentimun yang dikategorikan ke dalam lima kelas: Bercak Daun Bakteri, Penyakit Bulai Berbulu, Daun Sehat, Penyakit Mosaik, dan Penyakit Bulai Tepung. Metode yang diterapkan meliputi praproses (pengubahan ukuran, augmentasi, normalisasi), pelatihan model, pengujian, dan evaluasi menggunakan metrik akurasi, presisi, recall, dan skor F1. Model mencapai akurasi 88% pada data pelatihan, 84% pada data validasi, dan 81,50% pada data pengujian. Model yang telah dilatih kemudian diintegrasikan ke dalam aplikasi berbasis web menggunakan Streamlit untuk memfasilitasi klasifikasi interaktif. Hasilnya menunjukkan bahwa Jaringan Saraf Konvolusional (CNN) efektif dalam mengklasifikasikan penyakit daun mentimun secara otomatis dan dapat diterapkan sebagai solusi teknologi di bidang pertanian.

Development of a Web-Based Bus Rental Information System at PT. Dzakki Buana Tour

Faiz Abdulfattah — 2025-09-13

The development of information technology has had a significant impact on various sectors, including services, operational efficiency, and data management. Information technology enables business processes to be conducted more quickly, accurately, and in an integrated manner, while also reducing the risk of errors from manual record-keeping. PT Dzakki Buana Tour, a company providing tourist bus rental services based in Pekanbaru, still carries out its business processes manually using tools such as Word, Excel, and Spreadsheets. This often leads to data entry errors, scheduling conflicts, and delays in service. This study aims to analyze and design a web-based bus rental application for PT Dzakki Buana Tour. The result of this research is a system design that includes client data management, rental processing, departure scheduling, and bus and driver management. The system design is illustrated using UML modeling tools such as use case diagrams, activity diagrams, and class diagrams. This system is expected to improve the efficiency and accuracy of the bus rental operations at PT Dzakki Buana Tour.

Development of Virtual Reality Learning Media for Human Digestive System Materials in Junior High School

M.Rizky Maulana — 2025-09-13

The use of virtual reality (VR) technology is one of the innovative approaches in the development of learning media. This technology is able to present more real and interactive visualizations so that it makes it easier to understand complex materials such as the human digestive system and abstract concepts in science lessons. The use of technology-based media such as VR is important so that teachers and students can keep up with the times and take advantage of technological advances in the teaching and learning process, especially at SMP Muhammadiyah 1 Pekanbaru. This research aims to design and develop virtual reality-based learning media on human digestive system materials for grade VIII junior high school students. The research was conducted using the Research and Development (R&D) method using a 4D model which includes the stages of define, design, develop, and disseminate. The media developed has been validated by media experts and material experts, and tested on students. The results of the study showed excellent quality with a feasibility rate of 98% from media experts, 96% from material experts, and 93% from students. Thus, the virtual reality-based learning media developed was declared valid, very feasible, and effective to support the learning of digestive system materials.

Integration of Qdrant Vector Database and DeepSeek AI for Automated Chatbots on E-Commerce Applications

Azhar Abdurrafiq Sujana — 2025-09-13

Transformasi digital telah mendorong e-commerce untuk meningkatkan kualitas layanan pelanggan (customer service. Salah satu solusi yang ditawarkan yaitu munculnya teknologi chatbot sebagai program berbasis kecerdasan buatan yang mampu berinteraksi secara otomatis dengan pengguna. Penelitian ini mengembangkan chatbot dengan memanfaatkan Qdrant Vector Database sebagai vector untuk menyimpan dan mencari informasi berbasis konteks, dan OpenRouter API Key model DeepSeek AI sebagai akses chatbot. Chatbot ini dirancang untuk menjawab pertanyaan umum terkait informasi produk, stok, pengiriman, metode pembayaran serta pertanyaan umum lainnya dalam 24/7. Hasil penelitian menunjukan bahwa penerapan sistem ini membantu meningkatkan layanan otomatis kepada pengguna untuk mencari informasi terkait produk, pengiriman, pembayaran dan informasi umum lainnya, khususnya di luar jam operasional.

Optimasi algoritma deteksi spam email dengan BERT-MI dan jaringan dense

Florentina Yuni Arini — 2025-09-13

Email spam detection is a critical challenge in maintaining the security and efficiency of digital communication. This research proposes and evaluates an optimized pipeline for email spam detection by integrating Bidirectional Encoder Representations from Transformers (BERT) for feature extraction, Mutual Information (MI) for feature selection to reduce dimensionality, and a dense neural network for classification. The Lingspam dataset, consisting of 2893 emails (2412 ham and 481 spam), was used in the experiments with an 80% training and 20% testing data split. Text features were extracted using BERT (bert-base-uncased), resulting in a 768-dimensional embedding, which was then reduced to the 200 most relevant features using MI. A dense neural network model with a 256-128-64-32-1 neuron architecture was trained using the Adam optimizer, binary cross-entropy loss function, and techniques such as early stopping and class weights to handle class imbalance. Evaluation results on the test data demonstrated very high performance, achieving an accuracy of 99.14%, precision of 0.9596, recall of 0.9896, F1-score of 0.9744, and ROC-AUC of 0.9995. This approach indicates that the combination of BERT-MI with a dense network can achieve accuracy comparable to more complex methods, but with the potential for a simpler and more efficient architecture.

Gold Price Forecasting Based on Time Series Using the LSTM Deep Learning Architecture

Diva Arifal Adha — 2025-09-14

Gold is one of the most influential commodities in the global economy. Its high price volatility poses a significant challenge for investors, financial analysts, and policymakers in formulating effective strategies and making accurate decisions. Therefore, an accurate prediction method is needed to forecast future gold price movements. This study aims to forecast gold prices using a deep learning approach with the Long Short-Term Memory (LSTM) algorithm. The LSTM model is capable of learning long-term dependencies in time-series data, making it highly suitable for modeling complex and dynamic financial data. The data used in this study consists of daily historical gold prices obtained from reliable sources. A preprocessing phase was carried out to clean and normalize the data before training the model. Furthermore, this study compares the performance of the LSTM model with the Multilayer Perceptron (MLP) model to examine differences in prediction accuracy. Evaluation metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) were used to assess model performance. The results show that the LSTM model provides more accurate predictions compared to MLP, with lower error values and better model stability. In conclusion, the deep learning approach, particularly the LSTM model, can serve as an effective alternative for gold price forecasting and support data-driven decision-making in the financial sector.

Performance Analysis of K-Nearest Neighbors (KNN) and Random Forest Algorithms for Classification of Weather Conditions

Agim Sahrija Asha Yuda — 2025-09-19

The development of information technology has encouraged the use of machine learning algorithms in various fields, including in the analysis and prediction of weather conditions. This study aims to analyze and compare the performance of two machine learning algorithms, namely K-Nearest Neighbors (KNN) and Random Forest, in the classification of weather conditions based on historical meteorological data. The dataset used includes features such as rainfall, maximum temperature, minimum temperature, and wind speed, with target categories in the form of weather types such as rain, sunny, fog, drizzle, and snow. The process includes data pre-processing, feature scaling, training and test data sharing, and model training using the scikit-learn library. Performance evaluations are conducted using accuracy, precision, recall, and F1-score metrics. The results showed that the Random Forest model had higher accuracy (82%) than KNN (78%), with more stable performance in the majority class. However, both models experienced significant performance declines in minority classes due to data imbalances. The study recommends further optimizations such as class balancing and model parameter selection to improve the overall accuracy of weather classification.