Ida Nurhaida

This article presents a digit detection system for analogue electricity meters using the YOLOv9 algorithm integrated with EasyOCR. One hundred thirty-seven images from Plaza XYZ were used for training and testing, with an accuracy of 89.8%. The system enables real-time detection and export of meter readings, significantly reducing human error and recording time. The user-friendly interface allows technicians to upload images, detect digits, and export results to Microsoft Excel or CSV format. These findings demonstrate the model’s readiness for commercial deployment and advance the development of computer vision applications in energy management.

The differential distribution of cellular towers of Telkomsel, Indonesia’s largest mobile network operator, in Banten Province, Indonesia, poses challenges to network performance and service reliability. Therefore, we developed a novel hybrid framework that integrates geographic information systems, minimum spanning tree modeling, and signal-weighted k-nearest neighbor classification to assess tower utilization and signal coverage. Leveraging geospatial data from 110 Telkomsel cellular towers and 1000 simulated user nodes, it was found that 2.73% of towers were overloaded and 189 signal blank spots were identified in rural and topographically complex areas. By incorporating both spatial topology and signal strength sensitivity, the developed method outperforms conventional spatial or machine learning approaches in preserving spatial fidelity and supporting infrastructure planning. Despite the use of simulated user data, the framework demonstrates high scalability and adaptability for integration with real-time network performance metrics, enabling dynamic and location-specific telecommunication optimization.

Facial expressions are an important nonverbal indicator in measuring customer satisfaction levels. The objective of our research is to develop a real-time desktop application for detecting facial expressions as indicators of satisfaction using the Convolutional Neural Network (CNN) VGG19 architecture. The application development process employs the waterfall method, encompassing: needs analysis using the Japanese Female Facial Expression (JAFFE) dataset, comprising 213 grayscale images across seven emotional categories sourced from zenodo.org; followed by system design through format conversion, augmentation (rotation, flip), and mapping expressions into satisfaction categories, as well as Python-based application development with VGG19 model integration. The final stage involved black-box testing to verify functionality. Our findings resulted in a GUI-based desktop application using PySide capable of automatically detecting and displaying satisfaction levels in real-time without an internet connection. Validation results showed an accuracy of 89.39%, with the highest performance in the disgust and happiness classes and the lowest in the fear and sadness classes. The system successfully passed functional testing with expression detection functioning as designed, consistently displaying satisfaction mapping results. Our research demonstrates the potential of integrating CNN into efficient, objective, and flexible consumer satisfaction measurement across various operational conditions, making this application usable by businesses to monitor.

Skin cancer continues to show a significant global increase in incidence, and early detection remains essential to reducing mortality rates. Conventional diagnostic techniques such as biopsy are invasive, require considerable processing time, and are not always accessible, particularly in remote or resource-limited healthcare environments, indicating the need for an intelligent and efficient diagnostic support system. This study develops a lightweight Convolutional Neural Network (CNN) model designed to classify seven types of skin lesions using the HAM10000 dataset consisting of 10,015 dermatoscopic images. The preprocessing pipeline involved resizing, normalization, oversampling, and dataset splitting. The training process was conducted for a maximum of 40 epochs and concluded automatically at epoch 29 using early stopping to prevent overfitting. The experimental results demonstrated that the proposed model achieved an accuracy of 98%, and surpassed common pretrained architectures including ResNet50V2 (83%) and VGG19 (67%), with precision, recall, and F1-score metrics showing consistent performance across all lesion classes. The final trained model was integrated into the Dermascan web platform, enabling real-time automated lesion classification from user-uploaded images. These findings confirm that the lightweight CNN model offers a reliable, fast, and accessible tool for early skin cancer detection that can be beneficial for both clinical decision-support and wider public healthcare applications.

Enterprise Resource Planning (ERP) applications such as Odoo generally do not have predictive analytics capabilities for time series data and are limited to recording historical financial data. The limitations of ERP systems make companies dependent on traditional statistical models such as ARIMA, which often fail to capture complex non-linear patterns in financial data. However, the ability to accurately predict cash flow is crucial for strategic financial management in companies. This study aims to develop and evaluate a predictive model using a Long Short-Term Memory (LSTM) approach that is accurate and integrated into Odoo ERP. The research method includes designing a microservices architecture with FastAPI as a bridge between Odoo ERP, the predictive model, and prediction graph visualization. The LSTM model is evaluated by comparing it with the ARIMA model using 3,740 Daily cash flow data, with evaluation metrics MAE, MAPE, R2. System testing will use Black Box Testing and White Box Testing. The research results show that LSTM significantly outperforms the ARIMA model with an R2 evaluation of 0.8801 and an accuracy of 96.62%. The system testing results also yielded positive outcomes as the integration architecture runs stably and functionally. This research contributes by providing an Odoo ERP system that has predictive analysis capabilities with interactive graphical visualizations through Grafana, which helps companies make decisions effectively.

Women's participation in Science, Technology, Engineering, and Mathematics (STEM) is still low due to discrimination, gender stereotypes, and lack of access to equal career opportunities. This research analyzes public sentiment about gender equality in STEM fields using the Knowledge Discovery in Database (KDD) approach with the Long Short-Term Memory (LSTM) algorithm. The data consists of 1,200 tweets (2018-2024) collected through web crawling and processed using KDD techniques such as preprocessing, transformation, data mining and evaluation. The resulting LSTM model showed 86.25% accuracy, 88.18% precision, 82.20% recall, and 85.00% F1-score. Sentiment analysis showed support and appreciation for women in STEM (positive sentiment) and criticism of gender discrimination and stereotypes (negative sentiment). This study faced challenges in the form of data imbalance and the model's difficulty in understanding the Indonesian context. Our findings confirm the importance of policies that support gender equality and inclusive work environments. This research is expected to improve people's perception of gender equality and increase the representation of women in STEM fields, especially in Indonesia.

Identifying foot arch types is crucial for maintaining health and comfort. Flat foot arches can cause pain and discomfort, potentially interfering with activities such as sports. This research aims to develop an Artificial intelligence (AI)-based application to detect normal and flat foot arch types through X-ray images. The YOLOv8 model with bounding box is converted to TensorFlow Lite format to be integrated into a mobile platform through Android Studio. The application uses a waterfall model without maintenance, starting from the analysis of x-ray dataset needs, development and testing of the YOLOv8 model, conversion to TensorFlow Lite, design, black box testing, and application on Android devices. This application can only identify x-ray photos of the soles of the feet looking right and left. Confusion matrix application testing with 150 epochs shows performance with recall 86.2%, precision 77.1%, accuracy 83.3%, mAP50 94.9%, and mAP50-95 76.2%. Black box testing on mobile devices using datasets augmented with 45° horizontal shear and 90° rotation resulted in maximum identification accuracy compared to traditional methods such as the wet foot test. Traditional methods print the soles of the feet with an identification process that requires precision of the patient's standing position. This app detects flatfoot early, improving comfort in daily activities and sports.

This research aim is to propose a machine learning approach to automatically evaluate or categories hospital quality status using quality indicator data. This research was divided into six stages: data collection, pre-processing, feature engineering, data training, data testing, and evaluation. In 2020, we collected 5,542 data values for quality indicators from 658 Indonesian hospitals. However, we analyzed data from only 275 hospitals due to inadequate submission. We employed methods of machine learning such as decision tree (DT), gaussian naïve Bayes (GNB), logistic regression (LR), k-nearest neighbors (KNN), support vector machine (SVM), linear discriminant analysis (LDA) and neural network (NN) for research archive purposes. Logistic regression achieved a 70% accuracy rate, SVM a 68% accuracy rate, and neural network a 59.34% of accuracy. Moreover, K-nearest neighbors achieved a 54% of accuracy and decision tree a 41% accuracy. Gaussian-NB achieved a 32% accuracy rate. The linear discriminant analysis achieved the highest accuracy with 71%. It can be concluded that linear discriminant analysis is the algorithm suitable for hospital quality data in this research.

This research identifies the role of digital leadership in educational organizations. The analysis uses qualitative analysis. The research method used is SLR (Systematic Literature Review) with the PRISMA protocol through stages supported by the Publish or Perish and NVIVO applications. All supporting publications were searched through Publish or Perish and Dimension Database. The article search results found 543 related studies in 2020-2024, then filtered through the PRISMA protocol to 35 articles selected to answer the research questions. The results of the article are: 1) Sustainable digital leadership involves several essential aspects: agility, resilience, and adaptability. 2) There are 10 sustainable digital leadership competencies that digital leaders need to have: Focus on Vision, Repetitiveness, Communication and Collaboration, Flexibility, Resourcefulness, Risk-Taking and Recovery, Critical Thinking, Culture of Learning, Responsiveness, and Creativity and Innovation. 3) Future leaders must prioritize understanding digital change and assessing digital leadership competencies. This competency development can take the form of training, talent development, support from experts, and digital leadership assessments. 4) Essential to provide financial resources, infrastructure, work environment, and access to the latest learning technology. 5) The most vital aspect of digital leadership is how leaders collaborate and empower to realize the vision, implement change, and create a creative and innovative educational organizational environment.

The rapid advancement of digital technology has heightened the need for reliable methods to verify signature authenticity, a critical aspect of document and transaction security. This study uses a deep learning approach to develop a mobile application to verify the originality of paper and digital media signatures. The dataset comprises 1,060 signature images, including authentic and forged categories for both media types. The system employs the EfficientNetV2M model, trained with augmented data, to enhance robustness. Model evaluation demonstrates strong performance with an accuracy of 82.07%, a global precision of 81.31%, a global recall of 83.25%, and a global F1-score of 82.18%. The model is implemented in an Android-based mobile application, providing an intuitive interface for users to upload and verify signatures in real time. These results underscore the potential of EfficientNetV2M for mitigating signature fraud across various domains while highlighting areas for improvement, particularly in classifying paper-based signatures. Future work will focus on expanding the dataset and refining feature extraction techniques to enhance classification performance.