Journal of Applied Informatics and Computing

Knowledge Discovery Through Topic Modeling on GoPartner User Reviews Using BERTopic, LDA, and NMF

2025-01-10T03:27:26+00:00

Transportation and food delivery services are one of the driving sectors of the digital economy in Indonesia. The e-Conomy SEA 2023 report shows that the transportation and food delivery services sector experienced a decrease in GMV in 2023 by 8% from the previous year. The decline in GMV indicates a decrease in transaction value in the transportation and food delivery service sector. GoPartner is an application developed by GoTo to assist driver partners in carrying out various services in the gojek application which is one of the applications engaged in the transportation sector and food delivery services. Drivers as people who provide services directly to consumers are certainly one of the factors that influence customer behavior in using services. To find out the problems faced by drivers, this research conducts knowledge discovery through topic modeling on GoPartner application reviews using BERTopic, LDA, and NMF, each of these methods has a different approach. Based on the research results and the quality of the topics generated, BERTopic and LDA have better quality in analyzing GoPartner user reviews.

Application of Gated Recurrent Unit in Electroencephalogram (EEG)-Based Mental State Classification

2025-01-10T03:27:27+00:00

The classification of mental states based on electroencephalogram (EEG) recordings has recently gained significant interest in cognitive monitoring and human-computer interaction fields. Due to high signal variability and sensitivity to noise, correct classification is still tricky, even with advances in the analysis of EEG signals. Among deep learning models, Gated Recurrent Unit (GRU) models have established great potential for sequential EEG data analysis. The applications of the GRUs are less reviewed in tasks concerning classification cases of mental states compared to hybrid and convolutional models. Based on this paper, we will propose a method for developing a model based on the GRU network trained with raw EEG data in the classification tasks of mental states of concentration and relaxed conditions. We analyzed 400 EEG recordings taken from 10 subjects within a controlled environment and collected using the Muse EEG Headband. The mean, standard deviation, skewness, kurtosis, power spectral density, zero-crossing rate, and root mean square were extracted as statistical features from the raw EEG data. After parameter tuning, the GRU-based model achieved an excellent average accuracy value of 95.94% and also yielded precision, recall, and F1-scores within the range of 0.95 to 0.97 over 5-fold cross-validation. This shows that GRU works well in classifying mental states based on the EEG data.

Analysis of the Use of MTCNN and Landmark Technology to Improve the Accuracy of Facial Recognition on Official Documents

2025-01-10T03:27:28+00:00

A face recognition system consists of two stages: face detection and face recognition. Detection of features such as eyes and mouth is important in facial image processing, especially for official documents such as identity cards. To ensure identification accuracy, this research applies facial landmark extraction technology and MTCNN (Multi-Task Cascaded Convolutional Neural Network). The purpose of this research is to evaluate the accuracy of MTCNN in detecting facial features at the Department of Population and Civil Registration (dukcapil) Palu City, using facial landmarks and waterfall methods as an application development methodology. The evaluation results show that MTCNN has high face recognition accuracy and good positioning ability regardless of what GPU in use as long have right CPU and System Operation. In comparison, the Viola-Jones algorithm is effective for high-speed applications, while SSD offers balanced performance with GPU device requirements for optimal performance. While MTCNN proved to be effective, challenges still exist, such as false positives and false negatives, especially in poor lighting conditions and extreme poses. Image and camera quality, including resolution and facial expression, also affects detection accuracy. These findings suggest that the application of MTCNN can improve face recognition accuracy for official documents, although it requires addressing existing challenges. With this technology, it is expected that errors in facial recognition can be minimized, resulting in more reliable data that meets the standards for issuing identity documents. This research contributes to the development of a more accurate and efficient face recognition system for personal identification applications.

Comparative Performance Analysis of Optimization Algorithms in Artificial Neural Networks for Stock Price Prediction

2025-01-10T03:51:29+00:00

This study aims to enhance price prediction accuracy using Artificial Neural Networks (ANN) by comparing three optimization methods: Stochastic Gradient Descent (SGD), Adam, and RMSprop. The research employs a systematic approach involving the design, training, and validation of ANN models optimized by these techniques. Performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R Square are utilized to evaluate the effectiveness of each method. The results indicate that the Adam optimization method outperforms the others, achieving the lowest MSE of 0.0000503 and the lowest MAE of 0.0046, resulting in an impressive R Square value of 0.9989. Adam’s superior performance can be attributed to its adaptive learning rate mechanism, which effectively adjusts to the high volatility and noise characteristic of stock price data, enabling the model to converge faster and more accurately. In comparison, SGD produced a higher MSE of 0.0001208 and MAE of 0.0075, while RMSprop yielded an MSE of 0.0000726 and MAE of 0.0059. These findings highlight Adam’s ability to significantly enhance the predictive capabilities of ANN, particularly in dynamic and complex datasets, making it a preferred choice for this application. The novelty of this research lies not only in its comparative analysis of various optimization methods within the ANN framework but also in the exploration of unique ANN features and their application to a specific stock price prediction case study, providing deeper insights into the practical implications of optimization strategies. This study lays the groundwork for future research by suggesting the exploration of additional optimization algorithms and more complex neural network architectures to further improve prediction accuracy.

Coffee Beans Classification Using Convolutional Neural Networks Based On Extraction Value Analysis In Grayscale Color Space

2025-01-11T02:17:51+00:00

Coffee is a vital agricultural commodity, and precise classification of coffee beans is crucial for quality assessment and agricultural practices. In this study, we propose a methodology utilizing Convolutional Neural Networks (CNN) based on ResNet-101 architecture for coffee bean classification. The novelty of our approach lies in the integration of comprehensive feature extraction from grayscale coffee bean images, including mean, standard deviation, skewness, energy, entropy, and smoothness, with the transfer learning capabilities of CNN. Through this integration, we achieved exceptional classification performance, with the CNN model attaining accuracy, recall, precision, and F1-score metrics of 99.44% and 100% on the training data, and 100% on the testing data. These results underscore the robustness and generalization capability of our methodology in accurately classifying coffee bean types. While the dataset used in this study is experimental, the comprehensive feature extraction and the effectiveness of the CNN architecture suggest the potential for accurate classification of coffee bean types beyond the experimental data, provided the new data shares similar characteristics to the collected samples. For future research, we recommend exploring the integration of two transfer learning techniques within CNN architectures to further enhance coffee bean classification systems. Specifically, leveraging pre-trained CNN models as a foundation for feature extraction, while simultaneously fine-tuning specific layers to adapt to the nuances of coffee bean classification tasks, could offer improved model performance and scalability.

Colors and Texture Feature Extraction Using Learning Vector Quantization 3 Algorithm in Optimization of Beef Identification

2025-01-11T03:58:12+00:00

The Assessment Institute for Foods, Drugs, and Cosmetics of the Indonesian Ulama Council (LPPOM MUI) is responsible for conducting research, evaluations, and determining the halal status of products in accordance with Islamic teachings. In Indonesia, where religious diversity is prevalent, the halal certification process is crucial, particularly due to differences in the halal status of certain foods, such as beef and pork, across religions. One of the challenges in this process lies in ensuring a rapid and accurate determination of various types of meat, including beef, pork, goat, and buffalo, which currently tends to be time-consuming within the LPPOM MUI Halal Center. To address this issue, there is a need for a technological solution that can quickly and accurately identify different types of meat, thereby reducing consumer uncertainty when selecting halal products. This study aims to develop an Android-based application utilizing the Learning Vector Quantization 3 (LVQ3) method to facilitate the classification of meat types by analyzing patterns specific to beef, pork, goat, and buffalo. This system is expected to expedite the halal verification process, thereby supporting more efficient and accurate decision-making in the halal certification sector.

Development of Virtual Lab on Collision Dynamics Learning Object with Collision Algorithm Integration

2025-01-11T04:34:51+00:00

The objective of this study is to evaluate the efficacy of a Virtual Lab employing a collision algorithm in enhancing students' conceptual comprehension of collision dynamics, in comparison to traditional pedagogical approaches, within the context of physics education.The methodology employed in this study is as follows: The study employed an experimental approach, comprising a comparison between two groups: an experimental class that used the Virtual Lab, and a control class that utilised traditional teaching methods. Both groups were subjected to pre-tests to ascertain their existing level of understanding, after which post-tests were conducted to evaluate their knowledge after the instruction period. An independent t-test was employed to analyse the differences in post-test outcomes between the two groups.The results are as follows: The findings indicated a significant improvement in the experimental class's understanding, with an average increase from the pre-test to the post-test of 33.89%, in comparison to a 30.74% improvement in the control class. The results of the t-test demonstrated a statistically significant difference (t = 4.32, p < 0.05), indicating that the Virtual Lab was more effective in enhancing conceptual comprehension. In conclusion, the Virtual Lab, based on the collision algorithm, has been demonstrated to be an effective tool for teaching collision dynamics, offering a more interactive and engaging experience than traditional methods. This study highlights the potential of technology-based learning tools to enhance physics education and recommends further development of Virtual Labs with interactive features to increase accessibility and understanding in diverse educational environments.

Game-Based Learning for Mathematics Lesson on 3rd Grade Elementary School

2025-01-15T02:52:37+00:00

Integrating cutting-edge strategies to improve students' learning experiences has become increasingly important in the ever-changing world of education. This study investigates how third-grade students at SD Pius Purbalingga can benefit from using game-based learning as an instructional strategy to improve their mathematical education. The study focuses on how to hold children' attention and improve their knowledge of mathematics. The main subject of this study is the effectiveness of educational games in enhancing elementary school student's understanding of mathematics. A mathematics game was created to solve this problem by actively involving pupils and reiterating key mathematical ideas. This game-based strategy aimed to create an engaged and enjoyable learning experience for third-grade pupils with acceptable cognitive capacities. The findings suggest that students who played the math game significantly increased their involvement, comprehension, and memorization of mathematical ideas. This study adds to the growing evidence supporting using educational games as useful tools in mathematics instruction. The study's findings revealed increased academic performance among students, with male students experiencing a rise of 2.4% in their overall scores. In contrast, female students demonstrated a significantly higher increase of 8.5%, indicating a more pronounced advancement in their academic performance.

User Experience Evaluation of YouTube Website Using Eye Tracking Method

2025-01-15T04:53:17+00:00

YouTube is one of the most popular social media in Indonesia, with one of its features being the Clip Feature, which allows users to share 5-60 seconds video snippets, but many users still experience difficulty in accessing this feature. Based on a survey of more than 130 respondents, 60% were unaware of the Clip Feature, 85% had never used it, and 75% had difficulty finding its location in the YouTube interface. This research aims to evaluate the user experience in accessing the Clip Feature on the YouTube website using the Eye Tracking method, as well as analyzing user attention patterns. Through the RealEye.io tool, the results show that the quality of the test data is very good, with an average E-T data integrity value of 90.33% and gaze on screen of 89.73%. Heatmaps and gaze plot analysis show that respondents’ attention patterns tend to show confusion, especially when looking for the Clip feature. This is supported by the results of the attention & emotion graphs analysis, which overall show that the average attention level of respondents is at 0.318, with an increase in the emotion of surprise experienced by respondents more than the emotion of happy. Although the Clip Feature offers significant benefits, users still experience difficulties in accessing it, which results in a decreased user experience. This research is expected to provide new recommendations in improving the user experience of YouTube website, specifically to make the Clip feature more accessible and effective to use.

UX Analysis of the Virtual Tour 360 Application at Universitas Dr. Soetomo Campus

2025-01-15T07:05:00+00:00

This research investigates the effectiveness of the Virtual Tour 360 application implemented at Universitas Dr. Soetomo Campus, Surabaya, as a tool for enhancing prospective students' understanding and familiarity with campus facilities. Focusing on user experience (UX), this study evaluates key aspects such as the flow of the virtual tour, camera height for indoor and outdoor captures, image resolution and file size, and overall application size for online accessibility. User feedback highlights a high level of satisfaction, with 85.1% finding the application beneficial, especially on mobile devices, the preferred access method. The immersive 360-degree campus visualizations and user-friendly navigation have received positive responses, effectively providing a favorable first impression of the university. To further enrich user experience, optimizing mobile display quality and enhancing navigation features are recommended to offer a more comprehensive and interactive campus introduction.

Usability Evaluation of an E-Commerce Integrated with a Fan Community Platform Using Cognitive Walkthrough

2025-01-15T07:33:10+00:00

This study aims to evaluate the usability and user experience of Weverse Shop e-commerce app after integration into Weverse app using the Cognitive Walkthrough and Post Study System Usability Questionnaire (PSSUQ) methods. Cognitive Walkthrough is used to identify usability issues from an expert perspective, while PSSUQ is used to quantitatively measure user experience through three subscales: System Usefulness, Information Quality, and Interface Quality. Participants in this study ran 7 task scenarios relevant to the application features. Based on the analysis results, the average scores for the PSSUQ subscales were 2.99 for System Usefulness, 2.98 for Information Quality, and 2.87 for Interface Quality, with an overall score of 2.95. These results indicate that the application interface still needs improvement, especially in the aspects of navigation and information delivery. This research provides recommendations for improvements to usability elements to increase user satisfaction.

Evaluation of Telecommunication Customer Churn Classification with SMOTE Using Random Forest and XGBoost Algorithms

2025-01-15T08:34:20+00:00

Competition in the telecommunications industry, particularly among Internet Service Providers (ISPs), significantly influences customer churn, which negatively impacts revenue, profitability, and business sustainability. An effective approach to mitigate churn involves identifying potential churners early, enabling companies to implement strategic retention measures. However, predicting churn can be challenging due to the limited data available on churned customers. This study aims to predict customers likely to terminate or discontinue their subscriptions, focusing on addressing data imbalance using the Synthetic Minority Over-Sampling Technique (SMOTE). The dataset, sourced from Kaggle, comprises 21 attributes and 7,034 entries. The pre-processing phase includes data cleaning, feature encoding, and the implementation of Random Forest and XGBoost algorithms after data balancing with SMOTE. The findings reveal that the XGBoost algorithm achieves a prediction accuracy of 82%, outperforming Random Forest with 81%. Key factors influencing churn include Contract, TotalCharges, and tenure. The study concludes by emphasizing the significance of contract flexibility and the need to prioritize customers with high total costs or extended subscription periods to reduce churn rates. Future research is encouraged to investigate alternative methods for handling data imbalance and to explore advanced machine learning algorithms to further enhance prediction accuracy and the effectiveness of customer retention strategies.

Sentiment Analysis of Telegram App Reviews on Google Play Store Using the Support Vector Machine (SVM) Algorithm

2025-01-15T09:50:22+00:00

This study aims to analyze the sentiment of Telegram application reviews on the Google Play Store using the Support Vector Machine (SVM) algorithm. From a total of 14,700,000 initial reviews, a sampling technique was carried out to obtain 400 review data, which then went through the pre-processing stage to produce 345 review data to be classified. The SVM model used showed good performance with an accuracy of 81.16%, precision in the positive class reached 93%, recall in the negative class of 94%, and an average f1-score of around 81%. However, there was a discrepancy between the high rating and the content of the review, which highlighted the existence of high-rated reviews that contained criticism or vice versa. The confusion matrix analysis also showed some misclassification, where reviews should be categorized as positive sentiment but detected as negative, and vice versa. This research is expected to provide valuable feedback for Telegram application developers to improve the quality of service, although the results of this analysis have not been fully discussed in practice. The limitation of this study is that it only tested reviews that used Indonesian, which limited the scope of the findings to the context of local users.

Sentiment Analysis on Google Reviews Using Naïve Bayes, K-Nearest Neighbors, and Logistic Regression to Improve Novotel Services

2025-01-15T10:57:16+00:00

The application of artificial intelligence (AI) has been widely used in various industrial sectors, including the hospitality industry. One of the applications that is widely used in the hospitality industry is sentiment analysis. Sentiment analysis is carried out by analyzing feedback data from hotel guests or customers. The results of this sentiment analysis are important for decision makers to improve and improve their services. This study aims to obtain sentiment analysis results from Novotel hotel Google reviews based on machine learning by comparing three algorithms, namely Naïve Bayes, K-Nearest Neighbors (KNN), and Logistic Regression. The stages carried out in this study are data collection, data labeling, exploratory data analysis (EDA), data preprocessing, text representation, data sharing, modeling, model training, model evaluation, selection of the most accurate model, visualization of the most accurate model, interpretation of results and writing research reports. The dataset used was 1200 reviews, only 1190 reviews were used in the analysis. From the training results, the model produced by the Logistic Regression algorithm was the most accurate, namely 94.54% with unigrams (n = 1). Here are the results of each category, positive as many as 723 reviews (60.76%), negative as many as 218 reviews (18.32%), and neutral as many as 249 reviews (20.92%). Thus, most of the sentiment towards the service is positive, but some services need to be fixed and improved for customer satisfaction. The next research, the research area is expanded and the use of Deep Learning.

Comparison of EfficientNet-B0 and ResNet-50 for Detecting Diseases in Cocoa Fruit

2025-01-16T02:23:23+00:00

Cocoa is a plant that is very susceptible to disease. One of the diseases that often attacks cocoa is black spots on the fruit. Detecting diseases in cocoa fruit is usually done manually by experts, which has limitations in providing information and is very expensive. this study proposes a model for detecting cocoa fruit diseases based on deep learning, namely convolution neural networks (CNN). This study compares CNN architectures, namely EfficientNetB0 and ResNet50 because these two architectures are very popular. EfficientNetB0 is known to be efficient in utilizing model parameters and the ability to achieve high accuracy, while ResNet50 uses Residual block recognition which allows deeper and more accurate model training. The dataset used is 3344 healthy cocoa fruit images, 943 black pod rot images and 103 pod borer images. From this study, the results for the accuracy of both methods are equally superior with an accuracy of 96% while for the precision of the EfficientNetB0 architecture is superior to ResNet50 with a value of 95.7% while for recall and f1-score ResNet50 is superior with a recall value of 94.7% and f1-score 93.3%. Based on the Confusion Matrix, it can be seen that ResNet50 is able to predict pod borer accurately so it can be concluded that in this study ResNet 50 is superior to EfficientNetB0. However, ResNet50 requires more parameters than EfficientNetB0 so ResNet50 requires a very large amount of data and when using a small amount of data EfficientNetB0 is more suitable for use.

Clustering Time Series Forecasting Model for Grouping Provinces in Indonesia Based on Granulated Sugar Prices

2025-01-16T03:35:09+00:00

Clustering time series is the process of organizing data into groups based on similarities in specific patterns. This research uses the prices of granulated sugar in each province of Indonesia. According to USDA reports, sugar consumption in Indonesia in 2023 reached 7.9 million tons. On April 26, 2024, the price of granulated sugar peaked in the Papua Mountains at Rp29,320 per kg, while the lowest price was recorded in the Riau Islands at Rp16,460 per kg. The research aims to cluster provinces based on the characteristics of granulated sugar prices and to use forecasting models for each group. Two groups were formed based on the price patterns of granulated sugar over time. The provinces of Papua and West Papua are in group 2, while the other 30 provinces are in group 1. The best model developed using the auto ARIMA method is ARIMA (2, 1, 0), with a MAPE value of 2.36% for cluster 1, and ARIMA (1, 1, 1), with a MAPE value of 2.59% for cluster 2. These values are less than 10%, indicating that the models built using the auto ARIMA method for clusters 1 and 2 are suitable for forecasting.

Public Sentiment Analysis of the Free Meal Program: A Comparison of Naive Bayes and Support Vector Machine Methods on the Twitter (X) Social Media Platform

2025-01-16T05:13:07+00:00

The problems of nutrition, including stunting, remain a challenge in Indonesia. Therefore, Prabowo and Gibran launched the 2024 Free Meal Program, which provides free lunch to every school child as well as pregnant mothers. This research analyzes public sentiment towards this program using data from X with Naïve Bayes and Support Vector Machine (SVM) methods. The data was analyzed through crawling, preprocessing, labeling, and feature extraction using TF-IDF. The results showed a predominance of positive sentiment towards the program, with SVM performing better in sentiment classification, achieving 86.42% accuracy compared to Naïve Bayes with 67.9%. The findings can guide policymakers in improving the communication strategy and implementation of the Free Meal Program to make it more impactful for Indonesians.

Comparing Machine Learning Algorithms to Enhance Volumetric Water Content Prediction in Low-Cost Soil Moisture Sensor

2025-01-16T05:51:24+00:00

Measuring soil moisture is possible either with directly using gravimetric test or indirectly using soil moisture sensor. Direct measurements offer accuracy but are not efficient in field measurements. On the other hand, indirect measurement offers remote measurement that will facilitate the user but lacks in accuracy. This research aims to compare and identify the best machine learning model that can improve indirect measurement (soil moisture sensor prediction) using direct measurement (gravimetric test) as a response variable. This research uses linear regression, K-Nearest Neighbours (KNN) and Decision Tree models. The three models were then compared based on Root Mean Square Error (RMSE). The results suggested that KNN (0.02939128) had the smallest RMSE value followed by decision tree (0.05144186) and linear regression model (0.05172371).

Automatic Fish Feeding and Temperature Control System for Aquariums Based on Internet of Things (IoT)

2025-01-16T06:43:15+00:00

Keeping fish in aquariums has become one of the people's hobbies. An important factor in fish maintenance is the process of feeding and controlling the temperature of the aquarium. However, with various activities, fish care is often not carried out properly. This study develops an automatic system for feeding and controlling the temperature of the aquarium with goldfish as the test object. This study designs an automatic system to control the temperature and feeding in the aquarium using hardware such as a DS18B20 temperature sensor, load cell, and ultrasonic sensor. This system is controlled by ESP32 for reading sensor data and Arduino Uno for controlling the relay, cooling system, heater, and servo motor. ESP32 reads sensor data and sends it via MQTT to Node-red. Based on this data, the system regulates the temperature by activating the cooler (peltier and water pump) if the temperature is >28℃ and turning off the cooler when the temperature is <26℃. The heater is active if the temperature is <24℃ and stops when the temperature reaches 26℃. Feeding is carried out according to schedule, with servo 1 dropping feed into the load cell until the weight reaches the target weight. After that, servo 2 moves the feed into the aquarium. If the weight has not reached the target, servo 1 continues to be active. Based on the test, the average percentage of error in the temperature sensor is 0,08%, the weight sensor is 1.10%, and the ultrasonic sensor is 1.61%. This system successfully performs four times a day feeding and controls the temperature within the optimal range for goldfish, which is 24-28℃. The test results show that this system functions well and is in accordance with the research objectives.

Development of YOLO-Based Mobile Application for Detection of Defect Types in Robusta Coffee Beans

2025-01-16T07:11:52+00:00

Improving the quality of Robusta coffee beans is a crucial challenge in the coffee industry to ensure that consumers receive high-quality products. However, the identification of defects in coffee beans is still largely performed manually, making the process error-prone and time-consuming. This study aims to develop a YOLO-based mobile application to detect defects in Robusta coffee beans quickly and accurately. The method employed in this study is YOLO, a deep learning-based object detection algorithm known for its real-time object detection capabilities. The application was tested using a dataset of Robusta coffee beans containing various defects, such as broken, black, and wrinkled beans. The test results indicate that the application achieves high detection accuracy, with the black bean class achieving 95.3% accuracy, while the moldy or bleached bean class records the lowest accuracy at 62.2%. This application is expected to assist farmers and coffee industry stakeholders in improving the quality of Robusta coffee beans and enhancing the efficiency of the sorting process.

Lung X-ray Image Similarity Analysis Using RGB Pixel Comparison Method

2025-01-17T08:14:21+00:00

The high death rate caused by pneumonia and Covid-19 is still quite high. Based on data released by WHO, 14% of deaths in children under 5 years old are caused by pneumonia. One of the processes carried out to help the diagnosis process is to look at lung images using X-Ray images. To obtain information about normal lung X-Ray images, Pneumonia and Covid-19, calculations are carried out using the color difference in each pixel of the X-ray image. The calculation process will provide output in the form of numbers in units of 0 to 100. This is done to facilitate the process of identifying the similarity of each X-Ray image being compared. The research stages are carried out with stages starting from adjusting the image size, then by breaking down the pixel values of the two images being compared and the process of calculating the difference in value from each pixel with the same coordinates. After calculating a combination of 30,000 combinations using 300 x-ray images, the results obtained in the form of the level of similarity between normal x-ray images and pneumonia x-ray images are the highest with a similarity percentage of 80.06%. The combination of normal images and pneumonia images is 10,000 combinations using 100 normal x-ray images and 100 pneumonia x-ray images. Normal x-ray images and covid x-ray images have a similarity of 79.18%. The combination of normal images and covid images is 10,000 combinations. The combination uses 100 normal x-ray images and 100 covid x-ray images. Pneumonia x-ray images and covid x-ray images have the lowest similarity level of 78.87%. The combination of pneumonia x-ray images and covid x-ray images is 10,000 combinations. The data used in the combination are 100 pneumonia images and 100 covid images. From the test results, the information obtained was that Accuracy was worth 0.54, Precision was worth 0.54, Recall was worth 0.59 and F1-score was worth 0.56.

Analysis of Copy-move Manipulation in Digital Images using Scale Invariant Feature Transform (SIFT) and SVD-matching Methods

2025-01-17T08:55:02+00:00

In recent years, more and more data has been created in digital form, allowing for easier control over storage and manipulation thanks to technological advancements. Unfortunately, these advancements also bring with them many risks, especially those related to the security of digital files. One of the concerns of many organisations is digital forgery, as it is increasingly easy to create fake images without leaving obvious traces of manipulation. One form of image forgery known as ‘copy-move’ is considered one of the most difficult problems in forgery detection. In this case, a portion of an image is copied and pasted at another location in the same image to hide unwanted objects in the scene. In this paper, we propose a method that automatically detects duplication areas within the same image. Duplication detection is performed by identifying local characteristics of the image (key points) using the Scale Invariant Feature Transform (SIFT) method and matching identical features using the Singular Value Decomposition (SVD) method. The results obtained show that our proposed hybrid method is robust to geometric transformations and is able to detect duplication areas with high performance.

Utilization of Machine Learning for Predicting Corrosion Inhibition by Quinoxaline Compounds

2025-01-17T09:31:58+00:00

Corrosion is a significant issue in both industrial and academic sectors, with widespread negative impacts on various aspects, including economics and safety. To address this problem, the use of corrosion inhibitors has proven effective. This study explores the application of Machine Learning (ML) methods based on Quantitative Structure-Properties Relationship (QSPR) to develop a predictive model for the efficiency of quinoxaline compounds as corrosion inhibitors. By conducting a comparative analysis among three algorithms: AdaBoost Regressor (ADB), Gradient Boosting Regressor (GBR), and Extreme Gradient Boosting Regressor (XGBR), and optimizing parameters through hyperparameter tuning using Grid Search and Random Search, this research demonstrates that the XGBR model yields the most superior prediction results. The XGBR optimized with hyperparameter tuning using Grid Search achieved the highest R² value of 0.970 and showed the lowest RMSE, MSE, MAD, and MAPE values of 0.368, 0.135, 0.119, and 0.273, respectively, indicating high predictive accuracy. These results are expected to contribute to the development of more effective methods for identifying corrosion inhibitor candidates.

Optimization of Decision Tree Algorithm for Chronic Kidney Disease Classification Based on Particle Swarm Optimization (PSO)

2025-01-17T10:19:38+00:00

The body's most important vital organ is the kidney. The kidneys are responsible for maintaining acid and alkaline balance, regulating blood pressure, and filtering blood to prevent the accumulation of metabolic waste in the body. However, chronic kidney disease does not always show symptoms and signs but can progress to kidney failure. Algorithm-based predictive methods in data processing show great potential in the health field to predict various diseases, one of which is kidney disease. One of the techniques in data mining is classification. One of the classification algorithms in data mining that is often used to detect diseases is Decision Tree. In this study, it is expected that by combining these two methods, it will make a new contribution to the Decision Tree algorithm that is optimized with Particle Swarm Optimization (PSO) for the selection of relevant features, and improve the weaknesses in the model to improve more accurate predictions. By performing feature selection with the Particle Swarm Optimization (PSO) algorithm, it is shown that the use of Particle Swarm Optimization (PSO) can improve the accuracy and performance of the Decision Tree algorithm in the chronic kidney disease classification process. The accuracy of the Decision Tree algorithm with feature selection using Particle Swarm Optimization (PSO) is higher, reaching 0.967%, compared to the accuracy of Decision Tree without Particle Swarm Optimization (PSO) feature selection which is only 0.95%. This shows that Particle Swarm Optimization (PSO) is effective in selecting relevant features so that it can significantly improve model performance.

Apriori Algorithm Analysis to Determine Purchasing Patterns at Beleven Farma Pharmacy

2025-01-18T02:14:14+00:00

Beleven Farma Pharmacy is a place that provides medicines and other health products such as supplements, vitamins and also various health tests. As a newly established pharmacy, no innovations have been made to improve sales strategies. Analysis of purchasing patterns can produce information that helps pharmacies in determining product bundling recommendations as well as determining product layout. This research applies the a priori algorithm method and uses rapidminer tools to identify drug purchasing patterns from transaction data at the Beleven Farma pharmacy. The Knowledge discovery in database (KDD) method is used as a reference in the data processing process. Based on tests carried out by the author, the resulting rules are that if you buy hemaviton you will buy vice with 4% support and 91% confidence and if you buy amoxicillin you will buy paracetamol with 4% support and 64% confidence. Thus, the resulting information can be used to support decision making in determining marketing strategies so as to increase sales at pharmacies.

Implementation of the K-Nearest Neighbors (KNN) Regressor Method to Predict Toyota Used Car Prices

2025-01-18T03:00:40+00:00

The development of the automotive industry in Indonesia has experienced significant growth in recent decades, especially in the used car market segment. One of the used car brands that has high demand is Toyota, because it has a reliable reputation and quality. However, there are challenges that are often faced by sellers and buyers of used cars, namely in determining prices correctly and accurately. Incorrect pricing can be detrimental to one party, either the price is too high or too low. Prices that are too high can slow down the turnover of goods in the market. While low prices can cause sellers to experience losses. The purpose of this study is to help find good performance in determining the price of used Toyota cars. This study will use one of the Machine Learning methods, namely K-Nearest Neighbors Regressor. The KNN method is one method that can be used for classification and regression. In addition, this algorithm is a simple algorithm and can provide accurate prediction results based on its proximity to existing data. This study uses selected relevant features, namely model, year, kilometer, tax, mpg, and cc. The results of this study obtained MAE = 3.31686, MSE = 26.43640, RMSE = 5.14163, and R2-Score = 0.99501 using 90:10 data division and k = 1. This proves that KNN Regressor is an effective method in predicting the price of used Toyota cars. Therefore, the K-Nearest Neighbors (KNN) Regressor method is able to provide a fairly accurate price estimate with a minimal error rate.

Optimization of Urban Waste Collection Routes Using the Held-Karp Algorithm in a Web and Mobile-Based System

2025-01-18T04:49:37+00:00

In 2023, the Environmental Agency of Palu City recorded a total waste production of 97,492 tons, of which 10.4% was plastic waste. The Palu City Government operates a fleet of garbage trucks on a predetermined collection schedule. However, garbage bins frequently overflow before their scheduled pickup, resulting in extended waste accumulation and inefficiency. This study proposes a web and mobile-based system to enhance waste management by integrating bin condition reporting and shortest route calculation for collecting full bins. The Held-Karp algorithm is utilized to address the Travelling Salesman Problem (TSP) for determining optimal collection routes. The system was developed using Golang, Flutter, ReactJS, and a MySQL database. API functionality was validated using Postman, and overall system functionality was tested using the black-box method. A case study involving 8 test points (1 starting point, 10 waste collection points, and 1 endpoint) demonstrated that the proposed system reduces travel time by up to 21.74%, costs by 22.29%, fuel consumption by 21.16%, and distance traveled by 21.16% compared to conventional methods. These results highlight the potential of the system to significantly optimize waste collection operations and support sustainable urban waste management practices.

Optimization of Distribution Routes Using the Genetic Algorithm in the Traveling Salesman Problem

2025-01-20T09:40:06+00:00

Transportation plays a vital role in business operations, as it is essential for product distribution to maintain profitability. Optimizing distribution routes is crucial to reducing transportation costs, travel time, energy usage, and resource allocation while maximizing efficiency. Micro-entrepreneurs, particularly settled retailers, often face challenges in determining optimal travel routes, resulting in inefficiencies in product distribution. This issue is classified as a Traveling Salesman Problem (TSP), which involves finding the shortest possible route connecting several locations before returning to the starting point. To address this problem, this study applies a two-step approach: the greedy algorithm to provide an initial solution and the genetic algorithm for further optimization. The research employs both manual calculations and MATLAB 2018A software to solve the TSP. Results demonstrate that the optimized route reduces the travel distance by 1,260 meters compared to the initial solution, highlighting significant improvements in operational efficiency.

Analyzing the Impact of Artificial Intelligence on Student Learning: A Case Study of SMK Tri Arga 2

2025-01-20T10:04:53+00:00

The use of artificial intelligence (AI) technology in every aspect of life is a solution that provides an important contribution to the continuity of the wheel of life, not to mention in the world of education. In this digital era, AI has become an important partner and tool for students in completing assignments and assisting in carrying out their learning activities, especially in completing assignments given by teachers. The purpose of this study is to analyze the effect of artificial intelligence on learning, especially for students of SMK Tri Arga 2. The formulation of the research problem involves a description of the use of AI in the process of completing assignments, the benefits and challenges experienced by students, and the most dominant AI applications used in completing these assignments. The research method used is a quantitative descriptive method using a questionnaire, Likert Scale, which is created on Google Form and then sent to the Whatsapp Group and student personal chat. Using a random method. 105 students of SMK Tri Arga 2 participated in this study. the results of this study showed that 51.4% stated that they often use AI. Then 52.4% or 55 students answered that they often use AI in completing school assignments. These results certainly provide a new perspective on the role of AI in helping students complete the school assignments they have been given. It is also hoped that the results of this research can help Senior High Schools (SMA) and Vocational High Schools (SMK) in improving the quality of education and learning by integrating AI more effectively in their students' academic processes, improving supervision and regulations regarding the use of AI in completing assignments and making AI a companion tool while still paying attention to the ethics of plagiarism and the growth of students' skill development.

Energy Efficiency and Time Management through IoT-Based Automated Electrical Control System and Digital Prayer Schedule Integration

2025-01-21T04:15:39+00:00

This study aims to develop an Internet of Things (IoT)-based automated electrical control system integrated with a digital prayer schedule to enhance energy efficiency and time management in mosques. Using IoT devices, electronic equipment such as air conditioners, lights, and loudspeakers can be automatically controlled according to prayer times, reducing energy waste and minimizing fire risks due to negligence in operating electrical devices. The study employs a prototyping method, where a system prototype is tested to evaluate its performance and effectiveness. A machine-to-machine bridge communication model is implemented to integrate the digital prayer schedule with mosque electrical devices. The research results are expected to provide an environmentally friendly technological solution while improving mosque operational efficiency. However, technical challenges, such as integrating the system with older electrical infrastructure, limited internet connectivity, and user readiness to operate new technology, must be addressed. Thus, this study contributes not only to energy efficiency but also to improving mosque management quality through digital technology.

Analyzing Sentiment of SiCepat Express User Reviews

2025-01-21T07:23:53+00:00

The development of e-commerce in Indonesia has led to an increase in the number of users of product delivery services to deliver their customers' orders to their destination. SiCepat Ekspres is the number one fastest delivery service in Indonesia, besides JNE and JNT Express. The study aims to evaluate the performance of sentiment analysis methods in identifying and classifying sentiments related to SiCepat Ekspres. Data from Twitter media as many as 10,000 dataset records. The experimental results show that Random Forest with SMOTE is the best method, as it has the highest accuracy (91.10%), followed by improvements in precision, recall, and F-measure. SVM with SMOTE is in second place, with 90.50% accuracy and stable performance in other metrics. Naive Bayes with SMOTE shows improvement, but its performance remains slightly below Random Forest and SVM, with an accuracy of 88.80%.