Journal of Computer Networks, Architecture and High Performance Computing
Not a member yet
473 research outputs found
Sort by
Cracking Overtime: Unleashing Machine Learning at PT XYZ with Linear Regression, Neural Networks, and Random Forests
Get PDFExcessive overtime at PT XYZ is a significant issue for the organization. Besides the significant financial repercussions, they may also affect employee health and productivity. This research aims to forecast future overtime hours, facilitating strategic planning, mitigating excessive overtime, and developing more effective overtime policies. This study employs an overtime realization dataset encompassing many characteristics that influence overtime determinations. The dataset is partitioned into training and testing data to serve as inputs for the three predictive algorithm models: linear regression, artificial neural network, and random forest. The random forest model demonstrates superior performance, evidenced by a mean squared error (MSE) of 158.78, which is proximate to the actual value. The root mean squared error (RMSE) of 12.601 is lower than that of the other two models, indicating a reduced average prediction error. The mean absolute error (MAE) of 8.931 reflects the average deviation from the actual value, while the mean absolute percentage error (MAPE) of 0.336 indicates a prediction error of 34%. Furthermore, the coefficient of determination (R²) of 0.914 signifies that approximately 91.4% of the variation in overtime hours is accounted for, in contrast to the other models, which accounted for 78.8% and 79.6%, respectively. The results indicate that the random forest model demonstrates superior predictive accuracy compared to the other two algorithms, owing to its capacity to handle non-linear data and outliers. Consequently, the random forest model is advocated as the most efficacious method for forecasting the amount of supplementary working hours in the future
The Detection of Bullying Against Indonesian National Team Players Using Support Vector Machine
Get PDFDetection is a process to check or conduct an examination of something using certain methods and techniques. Detection can be used for various problems, for example in detection bullying, especially on social media, is a significant problem with negative impacts on mental health, especially for public figures such as Indonesian National Team players. This study aims to detect bullying comments on the Instagram platform using the Support Vector Machine (SVM) algorithm. The research dataset consists of 3,100 comments collected from the official Indonesian National Team account, which are classified into bullying and non-bullying categories. The data preprocessing stages include case folding, tokenizing, normalization, removing stopwords, and stemming. The processed data was analyzed using the Term Frequency-Inverse Document Frequency (TF-IDF) method for feature weighting before being classified using SVM with a linear kernel and Naïve Bayes. The results showed that SVM performed better with an accuracy of 89%, a bullying category precision reaching 93%, and a recall of 83%. Meanwhile, the Naïve Bayes method produced an accuracy of 79%, with a bullying category precision of 76% and a recall of 86%. The non-bullying category in Naïve Bayes has higher precision (84%) but lower recall (72%). Thus, SVM is proven to be more effective in detecting negative comments due to a better balance between precision and recall. However, challenges such as informal language variations and data imbalance remain obstacles in the development of this model. This study contributes to the development of cyberbullying detection technology and supports the creation of a healthier social media environment
Multiscale Facial Detection using RetinaFace Architecture with Loss Function
Get PDFFacial recognition technology has become increasingly prevalent in modern applications, from security systems to social media platforms. However, one of the most significant challenges in this field remains the accurate detection of faces across varying scales, orientations, and image qualities. Traditional face detection methods often struggle when faces appear at different sizes within the same image or when dealing with low-resolution imagery, leading to inconsistent performance that can compromise system reliability. The RetinaFace architecture emerges as a promising solution to address these multiscale detection challenges. By incorporating a Feature Pyramid Network (FPN), the system creates a hierarchical representation of features that enables effective detection of faces regardless of their size in the image. The FPN combines low-resolution, semantically strong features with high-resolution, semantically weak features, creating a robust feature pyramid that simultaneously captures facial characteristics at multiple scales. Context modules within RetinaFace further enhance detection capabilities by providing additional contextual information that helps distinguish faces from background noise and other objects. This comprehensive approach allows the system to maintain high accuracy even in challenging scenarios where faces appear small, partially occluded, or at unusual angles. The comparative analysis between ArcFace and SphereFace loss functions reveals important insights into optimization strategies for facial recognition systems. The experimental results on the WIDERFACE dataset demonstrate exceptional performance, with the RetinaFace-ResNet152-SphereFace combination achieving 94% accuracy. These findings highlight the importance of architectural choices and loss function selection in developing robust facial recognition systems capable of handling real-world deployment challenge
Machine Learning for Securing API Gateways : a Systematic Literature Review
Get PDFThe rapid growth of mobile banking has improved access to financial services but also introduced heightened cybersecurity risks, particularly due to vulnerabilities in API Gateways and limited user awareness of cyber threats. This study conducts a Systematic Literature Review (SLR) to explore how machine learning (ML) can address both technical and human-centric security challenges in digital banking. By reviewing sixteen peer-reviewed studies published between 2019 and 2025, the study identifies key ML techniques such as anomaly detection, behavior-based models, and deep learning architectures that are effective in detecting and mitigating API-based attacks. In parallel, the review examines ML applications aimed at enhancing user cybersecurity awareness, including personalized alert systems, user segmentation, and adaptive education mechanisms. Thematic synthesis reveals several challenges, including data availability and privacy, the interpretability of complex models, and integration with existing banking infrastructures. However, the study also highlights significant opportunities, such as the use of federated learning to preserve privacy, explainable AI to improve trust, and dynamic alert systems to prevent user fatigue. Based on the synthesis, a conceptual architecture is proposed to integrate ML-driven API threat detection and user education within mobile banking platforms. The findings provide valuable insights for both academic research and practical implementation, contributing to the development of intelligent, user-aware cybersecurity frameworks in the financial sector.Keywords: API Gateway Security, Cybersecurity Awareness, Machine Learning, Mobile Banking, Systematic Literature Review
COMPARISON OF RANDOM FOREST AND SUPPORT VECTOR MACHINE ALGORITHMS IN THE CLASSIFICATION OF DYSPEPSIA DISEASE
No full textFunctional dyspepsia remains a prevalent gastrointestinal disorder globally, with a higher burden in low- and middle-income countries such as Indonesia. Diagnostic challenges are exacerbated by limited healthcare infrastructure and a low ratio of gastroenterologists. Machine learning approaches offer a promising solution to enhance diagnostic consistency and accuracy in resource-limited settings. This study aims to compare the performance of the Random Forest (RF) and Support Vector Machine (SVM) algorithms in differentiating dyspepsia from gastroenteritis using Indonesian clinical data. A quantitative experimental method was applied using patient medical records, including gastrointestinal disease categories, vital signs, and symptom profiles. Data preprocessing was carried out by handling missing values through imputation and Min-Max scaling normalization. The dataset was divided into 80% training data and 20% testing data using stratified random sampling. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. Random Forest demonstrated superior performance on all evaluation metrics compared to SVM. RF achieved 86.5% accuracy, 86.0% precision, 85.0% recall, and 85.5% F1-score, while SVM achieved 83.2% accuracy, 83.0% precision, 81.0% recall, and 82.0% F1-score. The 3.3 percentage point improvement in accuracy and 4.0 percentage point improvement in recall are clinically significant. Random Forest proved more effective in dyspepsia classification, showing better handling of complex clinical data interactions and more reliable diagnostic performance. These findings support the implementation of an RF-based decision support system in Indonesian healthcare facilities to improve diagnostic consistency and patient outcomes
Sentiment Analysis on Cyanide Case After 'Ice Cold' Aired with NLP Method using Naïve Bayes Algorithm
Get PDFInformation technology is developing increasingly rapidly, and the reach of the Internet has expanded even to remote areas. The public increasingly uses social media as a source of information that discusses all aspects of people's lives. Social media has a vital role for most people, one of which is the news of the cyanide coffee case. The Cyanide Coffee case was discussed again by netizens after Netflix raised this case in a documentary film entitled Ice Cold, which made the public even more convinced of the irregularities of the case. Based on this, sentiment analysis is needed to extract comments to obtain public opinion information. The sentiment analysis aims to create a sentiment model to determine public comments on this case. Therefore, this research was conducted to find out and classify public sentiment on the Cyanide Coffee Case using the Natural Language Processing (NLP) method, which is a text preprocessing process followed by the tokenization stage. Data filtering was used using Indonesian Stopwords, and then normalization was continued using Porter Stemmer. In this study, data collection was carried out based on public comments on Ice Cold shows on the TikTok platform using TikTok Comments Scraper. The test results show that the classification using naïve Bayes obtained the results of 22 negative comments, 4052 neutral comments and 34 positive comments. The classification results of this study are 87% accuracy, 97.6% precision, 87% recall, and 91.9% F-Score
IMPLEMENTATION OF A SMART HOME BASED ON INTERNET OF THINGS USING CISCO PACKET TRACER
Get PDFA smart home is a home that uses information and communication technology, especially the Internet of Things (IoT), to increase comfort, efficiency, security and energy management. The main idea of ??a smart home is to connect various electronic devices and systems in the house so that they can interact with each other and with the occupants of the house automatically or through centralized control. Users can control devices in the home remotely via mobile devices or computers. This allows them to regulate the temperature, check home security, or control equipment even when they are not at home. Smart homes are often equipped with sophisticated security systems, including security cameras, motion sensors, and alarms. Residents can monitor and control this system via their smart devices. Smart homes can be a practical and innovative solution to improve the quality of life and optimize home management. However, it is important to consider the security and privacy issues as well as the costs associated with using this technology. Using this smart home really helps users to carry out activities calmly outside the home because it can monitor the electrical conditions at home. The device used to create a smart home simulation is using Cisco Packet Tracer
Prototype Of Moisture Content Meter In Grain Using Esp32 Based On Spreadsheet
Get PDFIn the process period after the rice is harvested, the rice is then separated from the stalk and referred to as grain which will then be dried. The dried grain aims to reduce the water content, in measuring the water content of the grain, an effective and efficient measurement and database storage tool is needed for users to find out which grain is suitable for processing and can determine the quality of the water content of the grain. The method used in this research uses the RnD (Research And Development) method. In this test using capacitive soil moisture sensor and using database storage in the form of google spreadsheet. The capacitive soil moisture sensor is also calibrated with conventional measuring instruments (Grain Moisture Meter) to find out whether the sensor works properly and accurately. The results in this test found that all components are able to work properly and show an error value <1, the sample reading data will be sent to the database on google spreadsheet so that users can find out the data records in real time and detail
Analysis of Gradient Boosting, XGBoost, and CatBoost on Mobile Phone Classification
Get PDFIn the ever-evolving landscape of mobile phone technology, accurately classifying device specifications is paramount for market analysis and consumer decision-making. This research conducts a comprehensive analysis of mobile phone specification classification using three prominent machine learning algorithms: Gradient Boosting, XGBoost, and CatBoost. Through meticulous dataset acquisition and preprocessing steps, including resolution normalization and price categorization, features essential for classification analysis were standardized. Robust cross-validation techniques were employed to assess model performance effectively. The study demonstrates the significant impact of normalization techniques on improving model performance across all algorithms and fold variations. CatBoost consistently emerges as the top-performing algorithm, followed closely by XGBoost, with Gradient Boosting displaying respectable performance. Notably, CatBoost consistently achieves the highest AUC values and accuracy scores, demonstrating superior performance in accurately classifying mobile phone specifications. These findings underscore the importance of preprocessing methods and algorithm selection in achieving optimal classification results. For mobile phone manufacturers, leveraging machine learning algorithms for effective classification can inform product development strategies, optimizing offerings based on consumer preferences. Similarly, for data analysts, employing appropriate preprocessing techniques and algorithmic approaches can lead to more accurate predictions and informed decision-making. Future research avenues include exploring advanced preprocessing methods, investigating alternative algorithms, and incorporating additional features or datasets to enrich the classification process. Overall, this research contributes to understanding mobile phone specification classification through machine learning methodologies, offering actionable insights for industry practitioners and researchers to address evolving market dynamics and consumer preferences
Implementing Histogram of Oriented Gradients to Recognize Crypto Price Graphic Patterns with Artificial Neural Network
Get PDFTechnical analysis stands as a pivotal strategy in analyzing graphic patterns to forecast future movements in crypto asset prices. However, comprehending numerous patterns poses a significant challenge for novice investors venturing into the investment realm. This study aims to facilitate investors in recognizing crypto price graph forms by classifying cryptographic price chart patterns. The dataset comprises images of seven types of crypto price graphic patterns obtained from the Kagle website, totaling 210 data points. A 70:30 training and testing data split is employed to ensure robust model evaluation. The study explores nine different Histogram of Oriented Gradients (HOG) parameter combinations for graphic pattern extraction. Leveraging the artificial neural network (ANN) classification method with parameter hyper tuning, the study assesses various HOG parameter configurations to optimize classification performance. The most optimal results are achieved with parameters Bin = 9, Cell Size = 16x16, and Block Size = 1x1, boasting an accuracy rate of 95.23%, precision of 95.55%, and recall of 95.23%. This classification approach streamlines the process for investors, enabling them to discern crypto price graph patterns effectively, thereby enhancing their investment decision-making capabilities in the dynamic cryptocurrency market landscape. By providing a structured method for pattern recognition, this study contributes to democratizing access to technical analysis tools, particularly benefiting novice investors seeking to navigate the complexities of cryptocurrency investment