Jurnal Nasional Pendidikan Teknik Informatika : JANAPATI

Ensembled Machine Learning Methods and Feature Extraction Approaches for Suicide-Related Social Media

Merinda Lestandy — 2024-07-27

Suicide is a pressing public health concern that affects both young people and adults. The widespread use of mobile devices and social networking has facilitated the gathering of data, allowing academics to assess patterns, concepts, emotions, and opinions expressed on these platforms. This study is to detect suicidal inclinations using Reddit online dataset. It allows for the identification of people who express thoughts of suicide by analyzing their postings. The method addresses and evaluates different machine learning classification models, namely linear SVC, random forest, and ensemble learning, along with feature extraction approaches such as TF-IDF, Bag of Words, and VADER. This study utilised a voting classifier in our ensemble model, where the projected class output is selected by the class with the highest probability. This approach, typically known as a "voting classifier," employs voting to forecast results. The results collected suggest that employing ensemble learning with the TF-IDF 2-grams approach yields the highest F1-score, specifically 0.9315. The efficacy of TF-IDF 2-grams can be determined to their capacity to capture a greater amount of contextual information and maintain the order of words.

Comparative Analysis of CNN Methods for Periapical Radiograph Classification

I Gusti Lanang Trisna Sumantara — 2024-07-27

Periapical radiographs are commonly used by dentists to diagnose dental problems and overall dental health conditions. The varying abilities of dentists to diagnose may be limited by their visual acuity and individual skills. To address this issue, there is a need for an application capable of computationally recognizing and classifying periapical radiographs. The commonly used computational method for image processing, specifically image recognition, is the Convolutional Neural Network (CNN) method. This study aims to create an application that can classify periapical radiographs and analyze the capabilities of the Convolutional Neural Network (CNN) method in this classification process. In general, periapical classification is divided into five types: Primary Endo with Secondary Perio, Primary Endodontic Lesion, Primary Perio with Secondary Endo, Primary Periodontal Lesion, and True Combined Lesions. The periapical radiograph classification process was tested using four CNN models: ResNet50v2, EfficientNetB1, MobileNet, and Shalow CNN. The evaluation of the CNN method utilized a confusion matrix-based technique to generate accuracy, precision, recall, F1-score and Weighted Average F1-score values. Based on the evaluation results, the highest accuracy value was achieved by EfficientNetB1 with 82%, followed by ResNet50v2 with 76%, MobileNet with 75%, and Shallow CNN with 71%.

Analyzing Hotel Owners Acceptance of Tiket.com Using Technology Acceptance Model

Baiq Dwi Zulianti Kurrotaa'yun — 2024-07-27

Lombok, a captivating island in West Nusa Tenggara, faces tourism challenges due to the COVID-19 pandemic, causing a decline in visits and revenue. Senggigi, known for its picturesque beaches, experienced a drastic drop in hotel occupancy. Focusing on post-pandemic recovery, the aim of this research is to investigate the adoption of Tiket.com applications by hotel owners in Senggigi, utilizing the Technology Acceptance Model (TAM) with a Partial Least Squares (PLS) approach. This research method is using a quantitative methodology, the study involves 50 respondents from 20 hotels, distributing questionnaires to explore perceptions of Online Travel Agent technology adoption based on TAM variables. The result of this research is the Partial Least Squares analysis indicates that perceived ease of use significantly influences perceived usefulness, emphasizing the importance of a user-friendly platform. While perceived usefulness alone may not directly impact usage intention, positive attitudes toward the system play a crucial role. The study recommends collaborative efforts between Tiket.com service providers and Senggigi hotel owners to enhance system adjustments, aligning with user needs and expectations. This research contributes valuable insights into technology's role in post-pandemic tourism recovery, providing a nuanced understanding of factors influencing the acceptance of Tiket.com within the Senggigi hospitality sector.

Requirements Engineering Quality: a Literature Review

Rosa Delima — 2024-07-27

Requirements Engineering Quality (REQ) has a large influence on the success of a software project. A systematic literature review (SLR) is conducted to get complete information about REQ. SLR reviewed 46 relevant publications from 2016 – 2022, sourced from three literature sources: Science Direct, Scopus, and IEEE. Based on the SLR, it is known that, generally, the artifacts processed for REQ are text requirements. The quality standards for REQ that are widely used are ISO/IEEE/IEC 29148 and IEEE 830, while the quality variables that are widely used are correctness, completeness, consistency, and defects/faults found in RE. A number of methods are used to perform automatic REQ. The most widely used method in publications is NLP. This is in line with most artifacts used in REQ, such as text requirements.

The Citizens Readiness for E-Government on The Jogja Smart Service (JSS) Application in Yogyakarta City

Lisa Sophia Yuliantini — 2024-07-27

This study aims to analyze the citizens readiness for e-Government on the Jogja smart service (JSS) application in the city of Yogyakarta. with indicators from Citizens’ readiness for E-Government (CREG), namely ICT Infrastructure, ICT Use, Human Capital, ICT Regulation, and Trust. The research method is quantitative, with questionnaire primary data totaling 100 respondents, and using Smart PLS software version 0.3. in conducting data analysis. The results show that ICT Infrastructure and ICT Use have a significant influence on the citizens' readiness for e-Government in the Jogja Smart Service (JSS) application. Whereas Human Capital, ICT Regulation, and Trust have no significant influence on the citizens' readiness for e-Government in the Jogja Smart Service (JSS) application. The limitation of this study is the number of respondents and the limited number of respondent variables are expected to be used as recommendations for further research.

Innovative Learning Model for Dharmagita Based on Telegram Chatbot

Ni Putu Utari Dyani Laksmi — 2024-07-27

In the digital era, instant messaging has become a vital aspect of people's daily lives, especially among the younger generation. This presents opportunities to utilize technology that can be integrated into instant messaging as a learning medium. This research innovates to develop a learning model for Dharmagita, also known as sacred Hindu songs, using a chatbot as a platform aimed at attracting the interest of the younger generation in studying Dharmagita as a cultural heritage. This chatbot was developed using the Rasa framework, which is founded on Natural Language Understanding (NLU). Based on the results of the User Acceptance Test, the Dharmagita Chatbot received a positive response from users. The chatbot model achieved an accuracy value of 86.7%, an F1-score of 88.4%, and a precision of 91.1%. These results underscore the effectiveness and reliability of the chatbot in facilitating learning and engagement with Dharmagita content.

Classification of Dog Emotions Using Convolutional Neural Network Method

Slamet Hermawan — 2024-07-27

The utilization of neural networks in dog emotion classification has great potential to improve the understanding of pet emotions. The goal is to develop a dog emotion classification system. This is important due to the lack of public ability to recognize and understand dog emotions. Neural networks able to create learning models can be used for decision-making, thus helping to reduce the risk of dangerous dog attacks. CNN itself is part of neural networks, where the CNN model has a higher accuracy rate of 74.75% compared to ResNet 65.10% and VGG 68.67%. Modeling using ROC-AUC shows the model's ability to distinguish emotion classes well. Angry has the highest AUC of 0.97, happy 0.93 and sad 0.96. While relaxed has the lowest AUC of 0.92. Classification report results show model has the highest precision and F1-Score values in angry class, while the highest recall value is in sad class.

Analysis of User Complaints for Telecommunication Brands on X (Twitter) using IndoBERT and Deep Learning

Valianda Farradillah Hakim — 2024-07-27

Tweeting on different official accounts is what users of Twitter (X) do most frequently. These tweets ranging from compliments to critiques. One of the official accounts that gets a lot of tweets from its customers is Telkomsel, an Indonesian telecom company. This study aims to find the maximum accuracy that can be obtained by combining CNN and Bi-LSTM algorithms with IndoBERT embeddings. A considerable accuracy level above 90% is demonstrated by the study, with CNN obtaining the greatest accuracy of 99% at a learning rate of 6*10^^-5, along with scores of 98%, 97%, and 97% for precision, recall, and F1 correspondingly.

Analyzing Technology Acceptance Model for Lombok Traditional Food Restaurant in GoFood Application

Joselina Rizki Bimantari — 2024-07-27

During the COVID-19 pandemic, Lombok Island, known for its stunning natural beauty, faced significant challenges in its culinary industry, resulting in a drastic decline in revenue and tourist visits. Embracing technological solutions, particularly through food delivery apps like GoFood, became pivotal in overcoming these obstacles. These apps not only sustained local restaurants during the pandemic but also preserved Lombok's distinctive cuisine, such as Sate Rembiga. Beyond pandemic resilience, GoFood played a crucial role in balancing global technological advancements and supporting daily activities. The aim of this research is to employ the Technology Acceptance Model (TAM) using Partial Least Square (PLS) approach to analyze the adoption of the GoFood application among owners of traditional Lombok cuisine restaurants, particularly focusing on the impact of the COVID-19 pandemic. This research method is designed to assess the extent of acceptance and adoption of GoFood among these restaurant owners during the pandemic, identifying key factors influencing their technological acceptance. The results of this research offer insights into the dynamics of technology acceptance within Lombok's culinary sector amid external changes such as the pandemic. In conclusion, understanding these dynamics can inform strategies to enhance the utilization of food delivery apps in traditional culinary businesses, ensuring resilience and adaptation in the face of unforeseen challenges.

Improving Image Retrieval Performance with SCS and MCS Clustering Techniques

Ulul Fikri — 2024-07-27

This paper presents two methods, Single Cluster Search (SCS) and Multiple Cluster Search (MCS), aimed at enhancing image retrieval performance on the Corel1k, Corel5k, and Corel10k datasets, which has a wide variation of images. The Multi Texton Co-Occurrence Descriptor (MTCD) method is used for feature extraction, and the K-Medoids and DBSCAN methods are used for dataset clustering. The clusters are then ranked based on the distance of their medoids to the query image. The most relevant images are retrieved from the highest-ranking clusters. SCS selects the cluster with the highest ranking as the search area and expands the search area to the next ranking cluster if the number of images is less than 6, which is the desired number of retrieval results. MCS merges several clusters with the highest ranking and combines clusters as the search area. Both methods are evaluated using several metrics, such as AP, MRR, and retrieval time. The results are also compared with the original method, which does not use clustering (the query image and the dataset are only extracted with MTCD, and their distance is calculated). The findings indicate that both methods improve the retrieval time. In Corel1k, the SCS method reduces the time complexity by 0.001s, while the MCS method, although not surpassing the original method, still shows potential. In Corel5k, both methods reduce the time complexity by 0.052s in the SCS method and 0.015s in the MCS method. In Corel10k, both methods reduce the time complexity by 0.122s in the SCS method and 0.058s in the MCS method, compared to the original method. These results have practical implications for improving image retrieval efficiency. The paper discusses the reasons behind these results and suggests possible directions for future research.

Electronic Payment Threats and Security: A Systematic Literature Review

Amelia Citra Dewi — 2024-07-27

In the emerging field of electronic payment systems, security challenges have become a major concern. This research addresses a comprehensive understanding of mitigation strategies for these threats. Through systematic literature analysis, we investigated the security vulnerabilities in electronic payment processes and discovered the latest blockchain technology as a strengthened security framework. Our findings reveal that while encryption and authentication provide the foundation of security, the integration of blockchain technology offers an unprecedented level of transaction integrity and transparency. This research not only highlights the urgent need for electronic payment security measures but also highlights the potential of blockchain and machine learning as transformative solutions. The implications of our research indicate an important shift in payment systems towards more secure and resilient electronic systems, paving the way for future research to explore the integration of cutting-edge technologies in combating ever-evolving cyber threats by leveraging blockchain technology, quantum computing and machine learning.

The Effectiveness of Augmented Reality Technology in Mathematics: A Case Study of SMP Al Azhar Plus Bogor

Ragil Siti Sholehah Siti — 2024-07-27

The field of mathematics education is rapidly advancing, particularly with the introduction of augmented reality (AR) technology as one of the tools used as an innovative learning medium. This research seeks to assess how the incorporation of AR influences students' grasp of mathematical concepts. This study is an experimental quantitative research using the experimental design method, with a population of 37 students from class IX of SMP Al Azhar Plus Bogor. The researcher obtained 30 students as samples based on purposive sampling technique, which were then divided into 2 groups: 15 students as the experimental group and 15 students as the control group. The experimental group underwent mathematics learning utilizing AR technology with 3D teaching materials that were prepared and accessible via smartphones through a barcode, covering topics such as congruent, similar, and spatial figures. Meanwhile, the control group followed conventional teaching methods using books as teaching materials with the same subject coverage. The study's results emphasize the substantial improvement in students' understanding of mathematical concepts through the effective utilization of AR. This improvement encompasses students' abilities to solve mathematical problems, retain conceptual memory, and actively participate in the learning process. Based on the statistical test results conducted, the experimental group obtained an average of 68.4153 or 68%, which falls into the category of moderately effective, while the control group obtained an average of 16.1508 or 16%, classified as ineffective. The Independent Sample Test yielded a Sig. (2-tailed) value of 0.000 < 0.05, indicating a significant difference in effectiveness between the experimental and control groups. Further data analysis indicates that the learning experience through AR not only provides a better understanding but also offers additional motivation to students, thereby increasing their interest in the subject of mathematics. Moreover, the study observes that a well-integrated instructional design within the curriculum, considering the context of AR usage, can contribute significantly to improved learning outcomes. The consequences of these discoveries strengthen the perspective that augmented reality (AR) is not just a successful educational instrument but can also offer a pleasurable learning encounter within the realm of mathematics education. The outcomes of this study play a substantial role in advancing more interactive approaches to teaching mathematics, with a specific emphasis on enhancing students' conceptual understanding. It is hoped that these findings can serve as a foundation for the implementation of AR in educational curricula as a strategic effort to enhance the quality of mathematics education and enrich students' learning experiences.

The Development of A Mobile-Based Area Recommendation System Using Grid-Based Area Skyline Query and Google Maps

Annisa — 2024-07-27

Choosing a location for a business or place of residence is an essential task in our daily lives. Typically, a location is considered favorable if it is in proximity to profitable facilities that enhance its value while being distant from facilities that diminish its value. However, conducting surveys in the field to identify desirable candidate locations is not always feasible. Factors such as high costs, inclement weather, and transportation limitations can hinder survey activities. This study aims to develop a mobile-based system for location selection using the Grid-based Area Skyline (GASKY) algorithm in conjunction with Google Maps. Google Maps is widely utilized for location-based decisions and is familiar to mobile application users. GASKY is employed for its capability to recommend locations based on user-provided information regarding desired facilities near the target location and facilities to be avoided, eliminating the need to input candidate locations from survey results. The outcomes of this study include a mobile-based application that utilizes the Google Maps API to create data collection modules. Mobile-based applications utilizing GASKY offer convenience, as they can be accessed by users anytime and anywhere.

Sign Language Recognition Based on Geometric Features Using Deep Learning

Eko Mulyanto Yuniarno — 2024-07-27

Sign language plays a crucial role in facilitating communication among individuals with hearing impairments. In Indonesia, the deaf community often rely on BISINDO (Indonesian Sign Language) to communicate amongst themselves. People who are unfamiliar with sign language will face difficulties. This research aims to develop a system for recognizing sign language using geometric features extracted from hand joint coordinates using Google's MediaPipe Hands framework. The dataset contains 12 common words, each recorded 30 times with 30 frames recorded for each instance. This will facilitate communication between deaf and hearing individuals. We conducted tests on LSTM- Geometric and CNN1D- Geometric models using geometric features, and on CNN-LSTM-Spatial and CNN1D-LSTM-Spatial models using spatial features. The results indicate that the LSTM model with geometric features achieved the highest accuracy of 99%. Geometric features have been shown to be more effective than spatial features for classifying sign language gestures.

Pneumonia Diagnosis Through Deep Learning: ResNet50v2 Model Implementation

Yufis Azhar — 2024-07-27

Pneumonia is a significant global health concern, particularly affecting young children and the elderly. It is a lung infection caused by bacteria, viruses, fungi, or parasites, leading to the alveoli filling with pus or fluid. This study addresses the challenge of accurately diagnosing pneumonia using chest X-ray images, a process traditionally dependent on the expertise of radiologists. The reliance on radiologists results in lengthy diagnosis times and high costs, particularly in regions with a shortage of medical professionals. This research presents a deep-learning approach to automate the classification of pneumonia using the ResNet50v2 model, which has been pre-trained on the ImageNet dataset. The dataset used in this study, obtained from the Guangzhou Women and Children’s Medical Center, comprises 5,856 images, with 1,583 normal and 4,273 pneumonia cases. The images were preprocessed and augmented to enhance the model's robustness. The proposed model achieved an accuracy of 94%, demonstrating its potential in clinical settings to assist in the rapid and reliable diagnosis of pneumonia. This study contributes to the growing body of research in medical image analysis by employing a pre-trained ResNet50v2 model. It highlights the importance of leveraging advanced machine-learning techniques to improve diagnostic accuracy and efficiency.

Improving Sentiment Analysis and Topic Extraction in Indonesian Travel App Reviews Through BERT Fine-Tuning

Oky Ade Irmawan — 2024-07-27

Abstract

The increasing use of the internet in Indonesia has an influence on the presence of Online Travel Agents (OTA). Through the OTA application, users can book transportation and accommodation tickets more easily and quickly. The increasingly rigorous competition is causing companies like PT XYZ to be able to provide solutions to the needs and problems of their customers in the field of online ticket booking. Many customers submit reviews of the use of the PT XYZ application through Playstore and Appstore, and it needs a technique to group thousands of reviews and detect the topics discussed by customers automatically. In this study, we classified reviews from Android and iOS applications using BERT that had been adjusted through fine-tuning with IndoBERT, as well as modeling topics using LDA to evaluate the coherence score of each sentiment. The result of the comparison of hyperparameter models for the most optimal classification is epoch 4 with a learning rate of 5e-5. The accuracy obtained is 0.91, with an f1-score of 0.74. In addition, testing was carried out to compare BERT with other traditional machine learning. The best performing algorithm was Logistic Regression using TF-IDF word embeddings, achieving an accuracy of 0.890 and an F1-score of 0.865. Therefore, it can be inferred that the accuracy achieved by the fine-tuned classification model of IndoBert is sufficiently high for application in the PT XYZ review classification. Using a coherence score, we found 29 positive topics, 6 neutral topics, and 3 negative topics that were considered the most optimal. This finding can be used as evaluation material for PT XYZ to provide the best service to customers.

IoT-Based Humidity Control System for Electrospinning

Luluk Arifatul Hikamiah — 2024-07-27

In recent years, electrospinning has become the most common and most widely used nanofiber manufacturing method. One parameter affecting the manufacturing process is the humidity parameter, which affects the morphology. The control tools that have been developed have not been equipped with a remote control and monitoring system. This research aims to develop a humidity control system that can be controlled and monitored remotely and is equipped with data recapitulation. This research used methods from literature studies on humidity control systems, manufacture and assembly of hardware and software, calibration, and performance tests. A humidity control device is produced using a NodeMCU microcontroller connected to the IoT-based Blynk application. This can be controlled and monitored manually with an LCD and Keypad or by using a handphone through the Blynk application, and the humidity value can be recapitulated with Google Sheets in real-time. The sensor used for reading this humidity value is a DHT 22 sensor which has an accuracy value of 99.72% and a precision of 98.81% for 45%-95% humidity. This IoT-based humidity control system device can automate the process of controlling, monitoring, and recapitulating humidity data in real-time.

Optimizing Machine Learning Performance with The Naive Bayes Classifier Process in Smart Farming

Made Yosfin Saputra — 2024-07-27

Indonesia is a country that relies heavily on the agricultural and plantation sectors to meet its needs for food and industrial raw materials. But farmers face challenges such as falling commodity prices and the negative impact of global warming, which has resulted in widespread drought. As a result, competition for water resources between the agricultural, industrial and household sectors is getting tighter, making it increasingly difficult for farmers to guarantee water supplies. The phenomenon of global warming has caused challenges in the current era. In Bali, although there is a method called “Subak” to manage rice field irrigation systems, it has not been fully implemented. To overcome this problem, a solution is needed that can automate water distribution based on soil moisture levels, temperature, light and air humidity. It uses machine learning techniques specifically using Naive Bayes Classifier to make real-time decisions regarding crop irrigation. The aim of this research is to increase the efficiency and effectiveness of crop irrigation in agriculture while reducing the impact of warming. The results of testing the scenario with orchid plants obtained an accuracy of around 80% and with general plants obtained around 80% which was tested every time 5 data were collected. Testing with a total of 84 training data and 26 test data. From the test results, an accuracy of 92.30769% was obtained.

An Improved Utility-Based Artificial Intelligence to Capture NPC Behaviour in Fighting Games Using Genetic Algorithm

Supeno Nugroho — 2024-07-27

In computer fighting games , the ability of players to play with Non-Player Characters (NPC) is essential. A poorly designed NPC causes poor player engagement due to predictable behaviour, thus leads to unsatisfactory playing experience. We propose utility-based AI selected by genetic algorithm to determine the utility functions of each NPC action. We applied ELO ratings (usually used in chess game) to determine fitness function. Utility-based artificial intelligence can deliver human-like NPC with varied decision-making and can employ many forms of function to calculate the AI utility value. Tests on chromosomes in each generation were also carried out to obtain different responses. The Pearson Correlation coefficient is used to obtain an analysis of the influence of each assessment variable. The simulation results verify the validity of our analysis and show that our scheme influences the satisfaction level of game users

Enhancing K-Means Clustering Model to Improve Rice Harvest Productivity Areas in Aceh Utara Using Purity

Sujacka Retno — 2024-07-27

To optimize the performance of the clustering process using K-Means, an optimalization approach employing the Purity algorithm is needed. This research was tested on a dataset of rice harvest productivity areas in Aceh Utara Regency by comprehensively analyzing the number of iterations and the DBI values produced by K-Means and Purity K-Means in clustering priority and non-priority rice production areas. This is in line with the efforts of the Regional Government to implement rice production intensification programs in Aceh Utara Regency. From the testing of Purity K-Means, an average of 5, 2, 2, 5, and 3 iterations were obtained from all tested datasets sequentially from 2019 to 2023. Meanwhile, from the testing of conventional K-Means, the average number of iterations obtained was 5.4, 4.8, 4.2, 5.6, and 3.8 iterations, sequentially. This indicates that the clustering performance conducted by Purity K-Means is better than conventional K-Means. The DBI values obtained from Purity K-Means for the entire dataset sequentially are 0.6781, 0.4175, 0.4419, 0.6182, and 0.4973. This value is lower compared to the DBI values obtained from conventional K-Means, which are 0.7178, 0.6025, 0.4971, 0.7222, and 0.5519, respectively. This also indicates that the validity level of the clustering results performed by Purity K-Means is higher than conventional K-Means.

A Comparative Study on the Impact of Feature Selection and Dataset Resampling on the Performance of the K-Nearest Neighbors (KNN) Classification Algorithm

I Gede Aris Gunadi — 2024-07-31

This study aims to evaluate the impact of dataset balancing and feature selection on the performance of the K-Nearest Neighbors (KNN) classification algorithm. The primary objective is to determine the effect of different training data balance ratios on classification performance. Additionally, the study analyzes the contribution of feature selection methods and data balancing to the overall performance of the classification algorithm. Three datasets (Titanic, Wine Quality, and Heart Diseases) sourced from Kaggle, were utilized in this research. Following the preprocessing stage, the datasets were subjected to three resampling scenarios with balance ratios of 0.3, 0.6, and 0.9. Feature selection was performed by combining correlation test values and information gain values, each weighted at 50%. The selected features were those with positive combined values of summation, correlation, and information gain. The KNN classification algorithm was then applied to datasets with and without feature selection. The results indicate that achieving a perfectly balanced ratio (ratio = 1) is not essential for improving classification performance. A balance ratio of 0.6 yielded results comparable to those of a perfect balance ratio. Furthermore, the findings demonstrate that feature selection has a more significant impact on classification performance than data balancing. Specifically, data with a balance ratio of 0.3 and feature selection outperformed data with a balance ratio of 0.6 but without feature selection.

QoS Analysis of Implementation Elastic WLAN Mechanism for Adaptive Bandwidth Management Systems in Smart Buildings

Gede Sukadarmika — 2024-08-01

The rapid growth of the Internet of Things technology has led to various innovative creations, However, effective management of data traffic generated by numerous sensors is essential to maintain network performance. This study develops and evaluates an adaptive Bandwidth Management System using Elastic WLAN to deal with the development of IoT system traffic so that network performance is maintained. Using Raspberry Pi as an Elastic WLAN device and a Hierarchical Token Bucket (HTB) running via Python script, this system manages Bandwidth allocation based on the number of visitors in the Smart Building. Evaluation was carried out in two rooms, comparing conditions before and after Elastic WLAN implementation. The results show that the implementation of Elastic WLAN improves network performance. This is indicated by improvements in the stability of upload and download rates, as well as very significant improvements in the Jitter and Latency parameters which are used as QoS parameters.