Journal of Computer Networks, Architecture and High Performance Computing
Not a member yet
473 research outputs found
Sort by
Twitter Sentiment Towards 2024 Jakarta Governor Candidates With Naïve Bayes Algorithm
Full text linkThis study aims to analyze public sentiment towards candidates for the 2024 Governor of DKI Jakarta through the Twitter platform, with a focus on classifying positive and negative sentiment. Along with the rapid development of social media, Twitter has become the main channel for people to voice political opinions. Sentiment analysis was conducted using the Naive Bayes algorithm to classify the sentiment of tweets collected through crawling techniques during the campaign period. The data used includes user tweets, with features such as frequently occurring words, popular hashtags, and discussion topics related to each gubernatorial candidate. The results showed that the Naive Bayes algorithm provided the best performance in classifying sentiment data in the period August 1 to December 26, 2024, with the highest accuracy rate reaching 75% at a data ratio of 90:10. This research also identified challenges in sentiment classification, such as the presence of new terms in test documents that are not recognized by the training model. The findings are expected to provide a clearer picture of public perceptions of gubernatorial candidates and contribute to the analysis of political sentiment on social medi
A Novel Privacy-Preserving Algorithm for Secure Data Sharing in Federated Learning Frameworks
Full text linkFederated Learning (FL) has emerged as a promising paradigm for the collaborative training of machine learning models across decentralized devices while preserving data privacy. However, ensuring data security and privacy during model updates remains a critical challenge, particularly in scenarios that involve sensitive data. This study proposes a novel Privacy-Preserving Algorithm (PPA-FL) designed to enhance data security and mitigate privacy leakage risks in FL frameworks. The algorithm integrates advanced encryption techniques, such as homomorphic encryption, with differential privacy to secure model updates without compromising the utility. Furthermore, it incorporates a dynamic noise-adjustment mechanism to adaptively balance privacy and model accuracy. Extensive experiments on benchmark datasets demonstrate that PPA-FL achieves a competitive trade-off between privacy protection and model performance compared to existing methods. The proposed approach is computationally efficient and scalable, making it suitable for real-world applications in healthcare, finance, and the IoT environment. This research contributes to advancing secure data-sharing practices in federated learning, fostering the broader adoption of privacy-preserving machine learning solutions
The implementation of the Random Forest Algorithm with Resampling and Without Resampling on the Hepatitis C Disease Dataset
Full text linkThis study evaluates the performance of Random Forest models for Hepatitis C classification using a dataset from Kaggle, focusing on addressing class imbalance through resampling techniques. We compare three approaches: baseline Random Forest without resampling, Random Forest with SMOTE+ENN (Synthetic Minority Oversampling Technique + Edited Nearest Neighbors), and Random Forest with SMOTE+OSS (Synthetic Minority Oversampling Technique + One-Sided Selection). Results show that the baseline model achieved high accuracy (0.9837) but failed to detect minority classes (e.g., suspect Blood Donor recall=0.00). SMOTE+ENN significantly improved performance, achieving perfect classification (precision=1.00, recall=1.00) for Hepatitis, Fibrosis, and Cirrhosis, while maintaining high accuracy (0.9919) and ROC AUC (0.9999). In contrast, SMOTE+OSS showed limitations in detecting Hepatitis (recall=0.00) and yielded lower precision for Fibrosis (0.44), indicating its undersampling approach may be too aggressive. The study highlights SMOTE+ENN as the most effective method for balancing class distribution and enhancing model robustness in medical diagnostics. These findings underscore the importance of selecting appropriate resampling techniques to improve minority class detection in imbalanced datasets, with implications for developing reliable AI-based diagnostic tools for Hepatitis C
Design of Smart Parking System Using Ultrasonic Sensor to Optimize Parking Lots On Campus
Full text linkMonitoring the availability of parking lots in the campus area is very important. This is related to the solution of the problem of limited parking spaces for four-wheeled vehicles. Existing parking spaces can be optimized by adding a vehicle detection device. This vehicle detection device uses ultrasonic sensors and its programming is based on the ESP32 Microcontroller. Sensitivity parameters measured are object detection distance, influence of other frequencies, influence of passing objects and range areas horizontally and vertically. In the research, the measurement results obtained are object detection distance up to 350 cm, the influence of other frequencies does not exist, passing objects can be detected by vehicle detection devices, range areas vertically up to 250 cm and horizontally up to 150 cm. Based on the test results, the distance reading by the ultrasonic sensor on the vehicle detection device is accurate. This measurement is in accordance with the specifications of the GH-311 type ultrasonic sensor used in the device
Arabic NLP: A Survey of Pre-Processing and Representation Techniques: Arabic NLP
No full textThe rapid growth of Arabic Natural Language Processing (NLP) has underscored the vital role of upstream tasks that prepare raw text for modeling. This review systematically examines the key steps in Arabic text pre-processing and representation learning, highlighting their impact on downstream NLP performance. We discuss the unique linguistic challenges posed by Arabic, such as rich morphology, orthographic ambiguity, dialectal diversity, and code-switching phenomena. The survey covers traditional rule-based and statistical methods and modern deep learning approaches, including subword tokenization and contextual embeddings. Special attention is given to how pre-trained language models like AraBERT and MARBERT interact with pre-processing pipelines, often redefining the balance between explicit text normalization and implicit representation learning. Furthermore, we analyze existing tools, benchmarks, and evaluation metrics, and identify persistent gaps such as dialect adaptation and Romanized Arabic (Arabizi) processing. By mapping current practices and open issues, this review aims to guide researchers and practitioners towards more robust, adaptive, and linguistically-aware Arabic NLP pipelines, ensuring that the data fed into models is as clean, consistent, and semantically meaningful as possible
Sentiment Analysis on TikTok Discourse Surrounding the 2024 North Sumatra Gubernatorial Election Using Support Vector Machine Algorithm
Full text linkThis study aims to analyze public sentiment towards the 2024 North Sumatra gubernatorial election by leveraging social media data, specifically TikTok, which has become a major platform for political discourse in Indonesia. The two competing candidate pairs, Bobby Nasution–Surya and Edy Rahmayadi–Hasan Basri, have sparked widespread online discussions that range from enthusiastic support to harsh criticism. These interactions have a significant impact on public opinion formation and may influence electoral outcomes. To address this phenomenon, this research implements a sentiment classification model using the Support Vector Machine (SVM) algorithm with a polynomial kernel, known for its effectiveness in handling high-dimensional textual data. A total of 2,100 TikTok comments were collected using scraping techniques via Python. The data then underwent several preprocessing stages, including case folding, cleaning, normalization, tokenizing, slangword removal, stopword removal, and stemming. Feature extraction was conducted using the TF-IDF method, followed by lexicon-based sentiment labeling into positive and negative classes. The classification model achieved an accuracy of 82%, with a positive sentiment precision of 0.81, recall of 0.96, and F1-score of 0.88. For negative sentiment, the precision was 0.86, recall 0.51, and F1-score 0.64. These findings indicate that the model performs well in identifying explicit positive sentiments but faces challenges in recognizing complex negative expressions such as sarcasm or implicit criticism. The results provide valuable insights into digital political behavior and demonstrate the potential of machine learning-based sentiment analysis as a tool for monitoring public perception in real time during elections
UI UX Design of Waste Sorting Website-Based Application Applying the Design Sprint Method Case Study Palmerah West Jakarta
Full text linkPublic awareness that is increasingly paying attention to environmental problems requires public awareness to manage waste in a more efficient way. Due to the lack of understanding and appropriate methods for sorting waste, solutions different from conventional methods must be sought. Technology has become an important part of innovation in management. User Interface (UI) and User Experience (UX) are increasingly considered important in meeting user needs due to new interaction patterns with technology. The aim of this research is to understand and meet user needs in waste sorting. Therefore, the author has found a solution by creating a UI/UX design for an interactive waste sorting application. This study was conducted using the design sprint method which consists of five steps: understand, differentiate, decide, create a prototype, and verify. The results of tests carried out using the system usability scale method produced a score of 75, which shows that this application makes it easier for users to sort waste more efficiently and environmentally friendly
Prediction of the Number of Patient Visits in a Psychiatric Hospital Prof. Dr. M. Ildrem Using Naive Bayesian Algorithm
No full textThis study was conducted to predict the number of patient visits at Prof. Dr. M. Ildrem Mental Hospital using the Naive Bayes algorithm, which is relevant given the increasing need for global mental health care. The main problem of this study is the difficulty in managing hospital resources efficiently due to unpredictable fluctuations in the number of patient visits. The research aims to apply the Naive Bayes algorithm to predict the number of patient visits and evaluate their performance. The method used is a naïve Bayes algorithm with systematic steps including historical data collection, data preprocessing using LabelEncoder, and dividing the dataset into training data and test data (80:20) where the training data totals 1331 data and the test data has 333 data. The Naive Bayes model is built and tested with metrics such as accuracy, precision, recall, and F1-score. The results of the study based on confusion matrix analysis, the model achieved an accuracy of 0.8108108108108109 or 81%, a precision of 0.8206686930091185 or 82.07%, a recall value of 0.9926470588235294 or 99.26%, and an F1-score of 0.90 or 90%, which shows that this model is quite effective in predicting service units with the dominance of adolescent category patient data where it is concluded that this prediction model is able to provide accurate estimates of patient visits, supporting the management of hospital resources, and improving the operational efficiency of mental health services. This research is expected to help hospitals in planning facilities and workforce more effectively
Analysis of Detergent Inventory Stock at Luch Laundry Using the Linear Regression Method
Full text linkInventory stock management is an important aspect in the laundry business to ensure smooth operations and minimize costs. Laundry Detergent shortages or overstocks can cause service disruptions and unnecessary additional costs. Therefore, a method is needed that can help predict stock needs accurately, one of which is the linear regression method. The data used includes historical data on detergent use and other factors that influence demand over several time periods. Through linear regression analysis, a predictive model can be built to estimate detergent needs in the future, so that stocks can be managed more efficiently. Research Method, namely the survey research method, is a research method carried out using surveys or direct data collection from Laundry Luch. The method/algorithm used to analyze the data is the linear regression method. The aim of this research is to apply the linear regression method in detergent inventory stock and to carry out analysis using the linear regression method in detergent inventory stock. The research results from the data that have been collected show that the predicted stock of detergent supplies for Laundry Luch in January 2025, with an estimated total usage of 111 boxes of detergent and a target usage of 95 boxes of detergent, is 129 boxes of detergent. The research conclusion is that the linear regression method provides real benefits in supporting data-based decision making
Profile Matching Method In Improving Staff Performance In Asahan University
Full text linkAsahan University (UNA) is one of the private universities in North Sumatra which has Institution B accreditation in 2024 which is led by the Chancellor Prof. Dr. Tri Harsono, M.Si, in improving quality in the field of administration, UNA always requires staff to give good performance according to their expertise. Objective assessment of job placement will improve performance and motivate staff to perform well. However, the process of placing staff positions at UNA is only based on tenure so that the performance process is less than optimal. Objective: Implementing a system that can provide an assessment of positions in accordance with their expertise. 50 employee data, diploma data, training certificates, work performance, work experience, work period and discipline. Based on the analysis of the process of assigning staff positions to UNA, there are more than 10 staff who have different positions. Furthermore, the data is processed using the profile matching method. The processing stage is through the selection of criteria and then it is processed to get the results of the analysis. Followed by calculating the level of accuracy with the results of the analysis of the UNA staffing section. The results of the testing of this method were that there were 16 staff who were very suitable with their expertise, while 28 people were deemed suitable and 6 more staff were not. So that the level of accuracy is 88%. The assessment of the test results has been able to provide an assessment of the appropriate position. but it can already be recommended to assist the staffing of UNA in analyzing the positions that are currently running to improve the performance of staff in the UNA environment