Claim Missing Document
Check
Articles

Found 10 Documents
Search
Journal : JOURNAL OF APPLIED INFORMATICS AND COMPUTING

Comparison of EfficientNet-B0 and ResNet-50 for Detecting Diseases in Cocoa Fruit Maylianti, Ni Putu; Wijayakusuma, I Gusti Ngurah Lanang; Arta Wiguna, I Putu Chandra
Journal of Applied Informatics and Computing Vol. 9 No. 1 (2025): February 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i1.8868

Abstract

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.
Detection of Political Hoax News Using Fine-Tuning IndoBERT Jocelynne, Charlotte; Wijayakusuma, IGN Lanang; Harini, Luh Putu Ida
Journal of Applied Informatics and Computing Vol. 9 No. 2 (2025): April 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i2.8989

Abstract

Indonesia has experienced a surge in the spread of political hoax news, posing a potential threat to democratic and social stability. This study aims to develop a model for detecting political hoax news in the Indonesian language using IndoBERT, a language model optimized for Indonesian text. The dataset was sourced from Kaggle and comprises 20,928 factual news articles and 2,251 hoax news articles from major Indonesian media outlets, including CNN, Kompas, Tempo, and Turnbackhoax. The imbalance between factual and hoax news articles was addressed through undersampling, resulting in 1,302 samples for each class. The research stages include data collection, preprocessing, IndoBERT model training, and model evaluation. Results indicate that fine-tuning IndoBERT can detect political hoax news with an accuracy of 94.1% and an ROC AUC of 0.991, demonstrating high performance in accuracy and generalization capability. This research is expected to contribute to minimizing the spread of political hoax news in Indonesia and enhance media literacy among the public.
Comparison of Machine Learning Methods for Menstrual Cycle Analysis and Prediction Khairunisa, Mutiara; Putri, Desak Made Sidantya Amanda; Wijayakusuma, I Gusti Ngurah Lanang
Journal of Applied Informatics and Computing Vol. 9 No. 2 (2025): April 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i2.9076

Abstract

This study compares three machine learning methods—Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Decision Tree—for analyzing and predicting menstrual cycles. The dataset consists of 1,665 samples with 80 attributes encompassing information related to menstrual health. These methods were evaluated using accuracy, Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) metrics. The results show that LSTM achieved the highest accuracy (91.3%), followed by CNN (88.9%) and Decision Tree (85.2%). LSTM excelled in capturing complex temporal patterns in menstrual cycle data, while CNN effectively identified key patterns, and Decision Tree offered interpretability despite lower performance. This study concludes that LSTM is the most effective model for menstrual cycle prediction. The findings highlight the potential for improved accuracy in reproductive health tracking, with future research opportunities to incorporate additional variables such as hormonal history and lifestyle factors, as well as a focus on data privacy.
Aspect-Based Sentiment Analysis of Reviews for Pandawa Beach Using Naive Bayes and SVM Methods Putri, Made Ayu Asri Oktarini; Sumarjaya, I Wayan; Wijayakusuma, I Gusti Ngurah Lanang
Journal of Applied Informatics and Computing Vol. 9 No. 2 (2025): April 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i2.9083

Abstract

The presence of digital technology, especially online platforms such as Google Maps, has changed the way people search for information about tourist destinations, including reviews and ratings from previous visitors. Aspect-based sentiment analysis becomes a very useful tool to understand people's views and feelings towards a place or product based on the reviews given and identify aspects of interest to tourists visiting Pandawa Beach, by utilizing Naive Bayes and Support Vector Machine (SVM) methods. The main objective of this research is to identify sentiment patterns based on aspects such as attraction, accessibility, amenities, and ancillary. Data was collected and labeled according to sentiment and aspects, then processed using preprocessing techniques, extracted by bag-of-words method, and chi-square feature selection. The model evaluation results showed that SVM produced the highest F1-Score value of 79,625%, while the Naive Bayes method reached 73.29%.
Named Entity Recognition for Medical Records of Heart Failure Using a Pre-trained BERT Model Manurung, Mikael Triartama; I Gusti Ngurah Lanang Wijayakusuma; I Putu Winada Gautama
Journal of Applied Informatics and Computing Vol. 9 No. 2 (2025): April 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i2.9170

Abstract

This study aims to develop a Named Entity Recognition (NER) model based on a pre-trained BERT model for medical records of heart failure patients. The focus of this research is to classify essential medical entities from unstructured medical record texts. The classification covers four categories: objective data (patient identity, laboratory test results, and objective examination data), subjective data (patient complaints), prescriptions, and diagnoses (diagnosis codes and descriptions). The methodology employs Natural Language Processing (NLP) techniques using Transformer-based architectures, such as Bidirectional Encoder Representation from Transformers (BERT). The developed model is evaluated based on entity label prediction accuracy and medical entity classification performance. The results indicate that the BERT-based NER model performs well, achieving an entity prediction accuracy of 84.82%. Furthermore, the model effectively classifies medical entities from input texts in alignment with expected medical entities. This research is expected to contribute significantly to medical data management, assist healthcare professionals in clinical decision-making, and serve as a reference for the development of AI-based healthcare technology in Indonesia.
Intelligent Web-Based Application for Personalized Obesity Management Wijayakusuma, I Gusti Ngurah Lanang; Sudarma, Made; I Ketut Gede Darma Putra; Oka Sudana; Minho Jo
Journal of Applied Informatics and Computing Vol. 9 No. 3 (2025): June 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i3.9151

Abstract

Obesity is a serious global problem due to its association with various chronic diseases. This study explores the utilization of machine learning in particular deep learning technology to predict Body Mass Index (BMI) from individual photos to create an efficient solution for assessing obesity. Using the ResNet152 model and K-Fold Cross Validation, this application integrates filters on individual photos to improve prediction accuracy. The application was developed using React JS for the front end, PHP and MySQL for the backend and database management, and Python as the core of the machine learning system. The application that tested using blackbox method, to see all features is functioning and the web application prototipe is passed all the test scenario.
Dendritic ShuffleNetV2 Model for Alzheimer’s Disease Imaging Classification Riandika Fathur Rochim; I Gusti Ngurah Lanang Wijayakusuma
Journal of Applied Informatics and Computing Vol. 9 No. 4 (2025): August 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i4.9742

Abstract

This study investigates the integration of a dendritic neural model (DNM) into the ShuffleNetV2 architecture to enhance Alzheimer’s stage classification from MRI scans. The proposed “Dendritic ShuffleNetV2” retains the original network’s computational cost (0.31 GFLOPs) while incurring only a 1.6% increase in parameter count (from 2.48 M to 2.52 M) and achieves faster convergence (15 epochs versus 22 epochs). Experiments were conducted on a four‑class Alzheimer’s MRI dataset comprising Non‑Demented, Very Mild Demented, Mild Demented, and Moderate Demented categories. Compared to the baseline ShuffleNetV2, the Dendritic variant yielded an average accuracy improvement of 0.79%, with corresponding gains of approximately 0.8% in weighted precision, recall, and F1‑score. Confusion matrix analysis revealed persistent overlap between the Very Mild and Mild Demented classes, although overall discrimination—particularly for the majority and early‑stage classes—remained robust. Training stability was maintained without significant overfitting.
Eye Disease Classification Using EfficientNet-B0 Based on Transfer Learning Pratiwi Tentriajaya, I Dewa Ayu Pradnya; Wijayakusuma, I Gusti Ngurah Lanang
Journal of Applied Informatics and Computing Vol. 9 No. 4 (2025): August 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i4.9743

Abstract

This study focuses on developing and evaluating a deep learning approach employing EfficientNet-B0 based on transfer learning to classify retinal fundus images into four categories: Cataract, Diabetic Retinopathy, Glaucoma, and Normal. The model was trained using a retinal image dataset and demonstrated stable training performance, indicated by a consistent decrease in both training and validation loss without signs of overfitting. The training accuracy reached 92%, while the validation accuracy ranged between 94–95%. Model performance evaluation using a confusion matrix and classification report showed excellent classification results, particularly for the Diabetic Retinopathy class, with an F1-Score of 0.98. The Cataract and Normal classes also achieved high performance, with F1-Scores of 0.94 and 0.92, respectively. However, classification accuracy slightly declined for the Glaucoma class, which experienced some misclassification with the Normal class. Overall, the model achieved a classification accuracy of 94% on the test dataset, indicating good generalization capability. These findings suggest that the model holds strong potential for implementation in automated medical image-based diagnostic support systems. Nonetheless, performance improvement in classes with relatively higher misclassification rates is still required to ensure model reliability in clinical practice.
Implementation of Convolutional Neural Networks (CNN) for Breast Cancer Detection Using ResNet18 Architecture Siden, Hagia Sofia; Wijayakusuma, I Gusti Ngurah Lanang
Journal of Applied Informatics and Computing Vol. 9 No. 4 (2025): August 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i4.9746

Abstract

Early detection of breast cancer is crucial for improving patient survival rates. This study implements a Convolutional Neural Network (CNN) architecture based on ResNet18 using a transfer learning approach to classify breast ultrasound (USG) images into three categories: normal, benign, and malignant. The dataset, comprising 1,578 grayscale images collected from Baheya Hospital in Egypt, underwent preprocessing steps including image conversion, normalization, and augmentation. The ResNet18 model was fine-tuned using selective layer unfreezing to better adapt to the medical imaging domain. Evaluation was conducted using stratified 5-fold cross-validation and assessed with accuracy, precision, recall, F1-score, and AUC metrics. The best results were achieved by fine-tuning layer2, layer3, and the fully connected layer, yielding 95% accuracy, a macro F1-score of 0.93, and an AUC of 0.9906. The findings demonstrate that ResNet18, when properly fine-tuned with transfer learning, delivers high performance in breast cancer detection via ultrasound and holds strong potential as a reliable clinical decision-support tool.
Comparison of Online Gambling Promotion Detection Performance Using DistilBERT and DeBERTa Models Pratama, Halim Meliana; Wijayakusuma, IGN Lanang; Widiastuti, Ratna Sari
Journal of Applied Informatics and Computing Vol. 9 No. 6 (2025): December 2025
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v9i6.11293

Abstract

Online gambling promotions on social media have become a serious concern in Indonesia, where perpetrators use ambiguous and disguised language to evade detection. This study compares two transformer-based models, DistilBERT and DeBERTa, in detecting such content within Indonesian YouTube comments. Using a balanced dataset of 6,350 comments, both models were fine-tuned with optimized hyperparameters (learning rate 1e-5, batch size 32, 5 epochs) and evaluated through five-fold cross-validation. Results show that DeBERTa achieves superior performance with 99.84% accuracy and perfect recall, while DistilBERT achieves 99.29% accuracy. Error and linguistic analyses indicate that DeBERTa’s disentangled attention and Byte-Pair Encoding provide better understanding of non-standard and ambiguous language. Despite requiring higher computational cost, DeBERTa is ideal for high-accuracy applications, whereas DistilBERT remains suitable for real-time and resource-limited environments.