Indonesian Journal of Data and Science

[RETRACTED] Comparison of Parameter Estimation Methods in Weibull Distribution

Dler Najmaldin — Mon, 31 Mar 2025 06:25:15 +0000

The main objective of this study is to compare the parameter estimation methods for Weibull distribution. We consider maximum likelihood and Bayes estimation methods for the scale and shape parameters of Weibull distribution. While computing the Bayes estimates for a Weibull distribution, the continuous conjugate joint prior distribution of the shape and scale parameters does not exist and the closed form expressions of the Bayes estimators cannot be obtained. In this study, we assume that the scale and shape parameters have the exponential prior and they are independently distributed. We use the Lindley approximation and the Markov Chain Monte Carlo (MCMC) method to obtain the approximate Bayes estimators. In simulation study we compare the effectiveness of the parameter estimation methods with Monte Carlo simulations.

Improving Part-of-Speech Tagging with Relative Positional Encoding in Transformer Models and Basic Rules

Jerome Aondongu Achir, Abdukarim Mohammad, Mohammed Abdullahi — Mon, 31 Mar 2025 06:24:23 +0000

Part-of-speech (POS) tagging is a critical task in natural language processing (NLP), influencing the performance of downstream applications such as semantic parsing and machine translation. This study presents a novel approach to POS tagging by incorporating relative positional encoding within the transformer model. Unlike traditional absolute positional encoding, the proposed method leverages token dependencies more effectively, enhancing the transformer’s self-attention mechanism. The architecture integrates a rule-based module to correct misclassifications, further refining the results. Experiments on the Groningen Meaning Bank (GMB) dataset demonstrate the model's superiority, achieving an accuracy of 99.68%, a significant improvement over the 98.60% accuracy of models with absolute positional encoding. Additional metrics, including precision (0.92), recall (0.89), and F1-score (0.90), further confirm the model's effectiveness. The findings highlight the potential of relative positional encoding in improving contextual understanding and model performance, providing a robust solution for POS tagging tasks in NLP.

Application of Mamdani Fuzzy Logic in Identifying Postpartum Depression Risk

Agnes Nola Sekar Kinasih, Moh Hosen, Anik Nur Handayani — Mon, 31 Mar 2025 06:00:12 +0000

Postpartum depression (PPD) is a psychological condition that affects many mothers and can persist for up to a year after childbirth, causing long-term complications for both mother and child. This research aims to develop an early detection model for postpartum depression using the Mamdani fuzzy logic approach, which addresses the uncertainty and ambiguity of symptom reporting. By employing fuzzy logic, the model interprets subjective symptoms, such as sadness and irritability, and categorizes risk into Low, Medium, or High levels. The methodology involved data preprocessing, defining fuzzy membership functions and rules, and validating the model through testing in Python and LabVIEW. The system achieved consistent results with a deviation of ±1% between implementations, demonstrating its reliability for practical applications. This study provides a foundation for future research to improve early detection systems, offering a supportive tool for healthcare professionals to facilitate timely intervention and enhance maternal and child health outcomes.

Performance Comparasion of DenseNet-121 and MobileNetV2 for Cacao Fruit Disease Image Classification

Mon, 31 Mar 2025 00:00:00 +0000

Diseases in Cocoa fruit are one of the main factors that reduce Cocoa productivity in Indonesia. Early identification od this disease is very important to increase yieldys and cocoa quality. This research compares the performance of two Convolutional Neural Network (CNN) architectures, namely DenseNet-121 and MobileNetV2, in classifying cocoa fruit disease image. The dataset used consists of 8000 image divided into four classes: fruit rot (Phytophthora palmivora), fuit sucking pests (Helopeltis), pod borers (Conopomorpha cramerella), dan healthy fruit (Healthy). The model was trained using data augmentation techniques and evaluated based on the metrics of precision, accuracy, recall, F1-score. The research results show that DenseNet-121 has higher accuracy, namely 94.50 %, while MobileNetV2 only produces an accuracy of 93.88%. From the results of these two models, which produces better accuracy values, namely DenseNet-121, will be developed to create a simple cocoa fruit disease detection system to help farmers identify diseases in cocoa fruit.

Implementation of Ensemble Deep Learning for Brain MRI Classification in Tumor Detection

Rahmat Fuadi Syam — Mon, 31 Mar 2025 00:00:00 +0000

This study proposes an innovative ensemble deep learning framework for the automated classification of brain MRI images aimed at tumor detection. In this research, two state-of-the-art convolutional neural network architectures, ResNet18 and DenseNet121, were individually fine-tuned and trained using a dataset comprising 7,023 images categorized into four classes: glioma, meningioma, no tumor, and pituitary tumor. Extensive data preprocessing, including image resizing to 224×224 pixels, various data augmentation techniques such as random horizontal flipping and rotation, and standard normalization, was applied to improve data quality and enhance model generalizability. Both models demonstrated high validation accuracies of approximately 97.7% when evaluated separately. Moreover, an ensemble approach was implemented by averaging the softmax probability outputs from each model, which led to a superior validation accuracy of 99.36%. This ensemble method effectively combines the complementary strengths of ResNet18 and DenseNet121, thereby reducing misclassification errors and ensuring robust diagnostic performance. The promising results obtained in this study suggest that ensemble deep learning can significantly improve the reliability of automated brain tumor detection, offering a valuable tool for clinical diagnosis. Future research will focus on further optimizing the ensemble strategy through advanced hyperparameter tuning and the seamless integration of additional models to further enhance overall performance.

An Optimization Strategy for Reducing CO₂ in Livestock Farming with IoT Integration and Decision Support System Approach Using Linear Programming

Annisa Fikria Shimbun, Muhammad Arif Alfian, Agam Saka Jati, Edi Faizal — Mon, 31 Mar 2025 06:06:37 +0000

Improper waste management in livestock environments can increase CO₂ emissions, impacting animal health and environmental sustainability. This study integrates IoT technology and a Decision Support System (DSS) to optimize efficiency in livestock waste management by reducing CO₂ levels. The method involves data collection using IoT sensors to measure CO₂ levels, humidity, and temperature, and the application of a Linear Programming model to determine when to spray Effective Microorganisms (EM4) solution and the frequency of spraying needed, or when to conduct waste dredging and its frequency to optimize CO₂ reduction. This model successfully maximizes operational cost efficiency by optimizing CO₂ reduction in the livestock environment. The conclusion of this study is that the integration of IoT and DSS provides an effective and efficient alternative solution in livestock waste management, supporting environmental sustainability through optimized CO₂ reduction.

Implementation of Support Vector Machine Algorithm for Classification of Study Period and Graduation Predicate of Students

Sumiyatun, Yagus Cahyadi, Edi Faizal — Mon, 31 Mar 2025 06:07:27 +0000

Higher education is a cornerstone for human resource development and plays a pivotal role in the progress of any nation. This study aims to analyze the factors influencing the duration of study and the classification of graduation predicates for students in the Information Systems program at UTDI, utilizing the Support Vector Machine (SVM) algorithm. The dataset comprises 500 student records containing academic and demographic attributes, such as GPA, age, number of semesters, and graduation predicates. The research methodology includes several stages: data collection, preprocessing, feature selection, model training, and model evaluation. The findings reveal that the SVM model successfully classified the study duration with perfect accuracy (100%), achieving precision, recall, and F1-score values of 1.00 across all categories of study duration: short, medium, and long. For graduation predicate classification, the model achieved an accuracy of 95.18%, despite some challenges in distinguishing between the "Cum Laude" and "Very Satisfactory" categories due to slight overlaps in GPA values. Other evaluation metrics, including precision, recall, and F1-score, also demonstrated favorable results, although categories with limited data entries exhibited higher classification difficulties. The analysis indicates that GPA, age, and the number of semesters are significant factors influencing study duration, while GPA emerged as the dominant factor in determining graduation predicates. Further evaluations suggest that enhancing data distribution and increasing the number of entries in certain categories could improve the model’s performance. This research makes a substantial contribution to the application of machine learning in educational data analysis, with potential implications for improving academic decision-making and strategic planning in higher education.

Comparison of ResNet-50 and DenseNet-121 Architectures in Classifying Diabetic Retinopathy

Mon, 31 Mar 2025 06:13:02 +0000

Diabetic Retinopathy (DR) is a severe complication of diabetes that can lead to blindness if not detected early. However, early diagnosis remains difficult due to subtle initial symptoms. With advancements in technology, deep learning has become a reliable approach for DR classification through retinal image analysis. This study compares the performance of two CNN architectures, ResNet-50 and DenseNet-121, in classifying DR using 2,000 retinal images categorized into four classes: normal, mild, moderate, and severe. Models were trained using both Stratified K-Fold Cross Validation (k=5) and non-K-Fold approaches. Data augmentation techniques, including rotation, flipping, zooming, and translation, were applied to enhance model generalization. Training was optimized using the Adam optimizer with a learning rate of 0.001, ReduceLROnPlateau for adaptive learning rate adjustment, and a dropout rate of 0.2 to prevent overfitting. Results showed that ResNet-50 outperformed DenseNet-121, achieving 84% accuracy (without K-Fold) and 83% (with K-Fold), while DenseNet-121 reached 80% and 81%, respectively. ResNet-50 also yielded higher precision, recall, and F1-score across all classes. Overall, the combination of a robust architecture, proper regularization, and optimization significantly improved classification performance and reduced overfitting.

Evaluating Machine Learning Approaches: A Comparative Study of Random Forest and Neural Networks in Grade Classification

Subitha Sivakumar, Sivakumar Venkataraman — Mon, 31 Mar 2025 00:00:00 +0000

This study presents a comprehensive comparative analysis of Random Forest and Neural Networks for grade classification using a dataset of 2,392 high school students. The research involves meticulous data collection and pre-processing, including scaling of numerical features and encoding of categorical variables, to optimize model training. Both models were implemented and rigorously evaluated using standard performance metrics, such as accuracy, precision, recall, and F1-score, with particular attention to their ability to correctly classify students into various grade categories. The results indicate that while the Random Forest model achieved a slightly higher baseline accuracy and benefits from enhanced interpretability, the Neural Networks model demonstrated competitive performance following hyperparameter tuning. The findings underscore the trade-offs between model interpretability and the capacity to capture complex nonlinear relationships, offering valuable insights for practitioners seeking to deploy effective classification strategies in educational settings. Future work may explore ensemble approaches and advanced feature engineering techniques to further improve classification performance.

Churn Prediction in Credit Customers Using Random Forest and XGBoost Methods

Bagas Akbar Maulana, Nurtriana Hidayati — Mon, 31 Mar 2025 06:13:51 +0000

The increasing use of credit cards provides significant economic benefits to banks but also presents challenges in managing risks, one of which is customer churn. Churn occurs when customers stop using credit card services, potentially leading to financial losses and reduced revenue. This study aims to predict customer churn in credit card users using the Random Forest and XGBoost methods. The dataset comprises 5,000 customer records, with class imbalance consisting of 800 churn and 4,200 non-churn cases, addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Analysis indicates that customers aged 40–55 dominate the churn group, with factors such as marital status, education level, transaction count, and total transaction value contributing to churn. Model evaluation shows that XGBoost achieves 96% accuracy with a precision of 86%, recall of 89%, and an F1-score of 87% for the minority class after tuning, although accuracy did not improve compared to pre-tuning. In contrast, Random Forest showed an increase in accuracy from 95% to 96% after tuning, with the F1-score for the minority class improving from 85% to 88%. Both models demonstrated excellent performance, but XGBoost excelled in identifying the minority class. Thus, XGBoost is recommended as the best model for churn prediction in this study.

Comparative Analysis of Gradient-Based Optimizers in Feedforward Neural Networks for Titanic Survival Prediction

I Putu Adi Pratama, Ni Wayan Jeri Kusama Dewi — Mon, 31 Mar 2025 06:14:45 +0000

The Titanic survival prediction problem has served as a benchmark for testing machine learning algorithms, particularly for binary classification tasks involving tabular data. While numerous models have been applied to this dataset, Feedforward Neural Networks (FNNs), also referred to as Multilayer Perceptrons (MLPs), offer unique advantages due to their ability to approximate complex functions. This study investigates the performance of FNNs for survival prediction using the Titanic dataset, focusing on the impact of gradient-based optimisation algorithms. Eight optimisers—Batch Gradient Descent (BGD), Stochastic Gradient Descent (SGD), Mini-Batch Gradient Descent, Nesterov Accelerated Gradient (NAG), Heavy Ball Method, Adam, RMSprop, and Nadam—were systematically compared across three FNN architectures: small ([64, 32, 16]), medium ([128, 64, 32]), and large ([256, 128, 64]). To enhance stability and generalisation, the models employed binary cross-entropy loss, dropout, L2 regularisation, batch normalisation, and Leaky ReLU activation. A dynamic learning rate scheduler was implemented to optimise training by adjusting the learning rate during each epoch. Models were trained using an 80-20 train-test split over 50 epochs, with performance assessed using metrics such as accuracy, precision, recall, F1 score, and cross-entropy loss. Results showed that Adam achieved the highest accuracy of 82.6% with an F1 score of 0.77 on the medium architecture, demonstrating the best balance between performance and training time. RMSprop and Nadam also delivered competitive results, particularly in terms of precision and generalisation. Smaller architectures were faster to train but showed reduced accuracy, while larger architectures marginally improved performance at the cost of longer training times. The inclusion of a learning rate scheduler further enhanced convergence and reduced overfitting, improving generalisation to unseen data. This study provides a comparative analysis of gradient-based optimisers for FNNs applied to tabular datasets, offering insights into the optimal configurations for balancing accuracy, generalisation, and computational efficiency. These findings contribute to the growing body of knowledge on leveraging neural networks for structured data tasks.

Use of Machine Learning in Power Consumption Optimization of Computing Devices

Mon, 31 Mar 2025 06:16:37 +0000

High power consumption on computing devices is a big challenge in the digital era that relies on intensive computing. This research aims to optimize power consumption on computing devices using machine learning. The random forest algorithm was chosen as the main model due to its ability to handle non-linear data and interactions between features. The data used includes historical power consumption, workload, environmental parameters, and device configuration. The research stages include data collection, data pre-processing, model development, evaluation, and implementation. The results showed that the random forest model successfully predicted power consumption with an accuracy of 92% and an RMSE of 0.15. After going through the hyperparameter tuning process, the model is able to reduce the prediction error by 10%. Implementation of the model on computing devices successfully optimized power consumption with savings of 15-20% over various workload scenarios, without compromising device performance. In addition, this research contributes to the reduction of carbon emissions by 5 tons of CO2 per year if applied to 100 servers, as well as operational cost savings of up to $50,000 per year at scale.

Optimizing Javanese Numeral Recognition Using YOLOv8 Technology: An Approach for Digital Preservation of Cultural Heritage

Lukman Syafie, Huzain Azis, Fadhila Tangguh Admojo — Mon, 31 Mar 2025 00:00:00 +0000

This study examines the YOLOv8 technology in identifying Javanese script characters from handwritten text, aiming to support digital preservation initiatives for Javanese cultural heritage. The research is grounded in the need to develop efficient and effective methods for recognizing and digitizing traditional handwriting. YOLOv8, an advanced object detection model, is tested to assess its ability to classify Javanese script characters from digital images. The research process includes the collection of labeled handwritten data, training the model using this dataset, and evaluating the model's performance through confusion matrix analysis as well as F1-Confidence, Precision-Confidence, and Precision-Recall curves. The results indicate that YOLOv8 delivers high precision and recall, affirming its effectiveness in recognizing handwritten Javanese script characters. The study concludes by highlighting the significant potential of YOLOv8 not only in character recognition but also as a tool for digitizing historical documents and educational materials. Consequently, the implementation of this technology can enrich learning resources and support cultural preservation efforts. Further research is recommended to develop a more comprehensive dataset and integrate augmented reality technology to enhance the learning experience of the Javanese script.

A Comperative Study on Efficacy of CNN VGG-16, DenseNet121, ResNet50V2, And EfficientNetB0 in Toraja Carving Classification

Mon, 31 Mar 2025 06:18:53 +0000

Passura', or Toraja carvings, are a significant part of the culture in Toraja and have deep symbolic significance and beautiful patterns. Inspired by folklore and natural elements, these motifs capture the harmonious coexistence of nature, the divine, and humans. In order to classify Toraja carving motifs, this study uses Convolutional Neural Networks (CNN), specifically the architectures VGG-16, DenseNet121, ResNet50V2, and EfficientNetB0. Seven motif types were represented in the 700 photos that made up the dataset, which was split between 80% training and 20% validation data. Despite showing signs of overfitting, the EfficientNetB0 model had the greatest validation accuracy of 98%. ResNet50V2 was the most successful model for this job, exhibiting strong performance with 95.33% accuracy. These findings demonstrate how CNN architectures, with precise motif classification, can be used to preserve and record cultural heritage. This study advances the field of digital cultural preservation and lays the groundwork for further research using bigger datasets to improve model dependability and reduce overfitting.

Sugeno Fuzzy Logic for IoT-based Chicken Farm Drinking Water Quality Monitoring

Rosmasari, Didi Nur Rahmad , Anton Prafanto, Aulia Khoirunnita, Muh Jamil — Mon, 31 Mar 2025 06:19:44 +0000

Indonesia's poultry industry is growing, with chicken and eggs being the most consumed source of protein in the country. Broiler chickens, chickens raised for meat, are one type of livestock production that requires innovation in managing them. Broiler chicken growth can be influenced by several factors, one of which is chicken disease. Diseases in broiler chickens are usually caused by the environment. One of the environmental factors that affects chicken disease is drinking water. Water is one of the most important things for the chicken's body. The largest component in the chicken's body is water which reaches 60-85% of all parts of its body. Chicken drinking water should not have an acidity level that is too high or too low. The Environmental Protection Agency or EPA states that the standard pH of drinking water that is good for daily consumption is in the range of 6.5 to 8.5. The chicken drinking water used in this study used water from a former mining lake that was collected and left for several days. Based on interviews, checking the pH of the water still uses a manual tool, namely litmus (pH measuring paper). By using IoT technology, the author designed an IoT tool using a pH sensor and a turbidity sensor, the results of which will be read from both sensors and then processed by a microcontroller and analyzed using the Fuzzy Sugeno method to determine water quality. System testing was conducted using the Confusion Matrix method to evaluate performance with results of 96.76% accuracy, 97.52% precision, 98.79% recall, and 98.15% F-score. This tool is expected to help farmers monitor water quality in real time and provide more accurate recommendations regarding safe water conditions for broiler chickens.