Claim Missing Document
Check
Articles

Found 4 Documents
Search
Journal : journal of artificial intelligence and engineering applications jaiea

Optimization of Convolutional Neural Networks Using Resizing Techniques for Banana Leaf Disease Classification Kurniawan, Aldiyansyah; Purnamasari, Ade Irma; Pratama, Denni; Tohidi, Edi; Wahyudin, Edi
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 5 No. 2 (2026): February 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v5i2.1876

Abstract

Early and accurate identification of banana leaf diseases is essential for supporting digital agriculture, as visual symptoms often require rapid and reliable analysis. This study investigates the impact of three image resizing techniques squashing, letterboxing, and random resized crop on the performance of the MobileNetV2 architecture in classifying four categories of banana leaf images using the Banana Leaf Disease Dataset v4 consisting of 4,675 samples. The experiments were conducted using a transfer learning approach with an 80:10:10 data split, standardized normalization, and data augmentation. The results show that all resizing techniques achieved test accuracies above 92%. Squashing produced the highest accuracy and fastest training time, letterboxing demonstrated the most stable performance with the lowest validation loss, and random resized crop improved generalization to variations in object position. These findings confirm that resizing strategies significantly influence the stability and effectiveness of CNN models. Overall, MobileNetV2 proves capable of delivering accurate and efficient classification of banana leaf diseases when supported by an appropriate preprocessing pipeline. This study provides empirical evidence for developing image-based plant disease diagnosis systems within smart agriculture.
Application of Weighted Loss Function in Convolutional Neural Network for Acne Image Classification Abubakar Sidik; Purnamasari, Ade Irma; Pratama, Denni; Marta, Puji Pramudya; Wijaya, Yudhistira Arie
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 5 No. 2 (2026): February 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v5i2.1885

Abstract

Automated acne image classification using Convolutional Neural Networks (CNN) holds significant potential in dermatological diagnosis but faces a fundamental challenge of class imbalance. This phenomenon causes standard models to be biased towards majority classes and fail to recognize clinically important minority classes. This study aims to address this bias by applying a Weighted Loss Function to the EfficientNetB1 architecture. The research method employs a comparative experimental approach between two scenarios: the Baseline model (Standard Cross-Entropy) and the Proposed model (Weighted Cross-Entropy). The dataset consists of 5 acne classes with an imbalanced distribution. The results show that the Weighted Loss model significantly outperforms the Baseline model. Overall accuracy increased from 80% to 86%. The most significant improvement occurred in the minority class 'Papules', where the F1-Score surged by 0.10 points (from 0.71 to 0.81). It is concluded that the application of Weighted Loss Function effectively overcomes bias due to imbalanced data without the need for synthetic data augmentation, resulting in a fairer and more reliable model for clinical implementation.
Analysis of the Effectiveness of Manual Deployment and CI/CD Github Actions in the Braisee Application Seputra, Nenda Alfadil; Nurdiawan, Odi; Dikananda, Arif Rinaldi; Pratama, Denni; Kurnia, Dian Ade
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 5 No. 2 (2026): February 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v5i2.1916

Abstract

In the modern cloud-based software development ecosystem, the speed and reliability of the deployment process are critical elements. This study aims to evaluate the effectiveness of implementing Continuous Integration/Continuous Deployment (CI/CD) using GitHub Actions compared to manual methods for the machine learning API of the Braisee application hosted on Google Cloud Run. Using a quantitative approach with a comparative experimental design across ten testing iterations, this research measures deployment time efficiency, error rates, and system stability. The experimental results show a significant performance disparity, where the automated method based on GitHub Actions is considerably more efficient, with an average total duration of 111–167 seconds, reducing operational time by 40–60% compared to the manual method, which requires 297–364 seconds. In terms of reliability, the automated method achieves a 100% success rate with high consistency, whereas the manual method demonstrates substantial vulnerability to human errors such as mistyped project IDs and inconsistent image tagging. It is concluded that implementing CI/CD through GitHub Actions is a superior solution that improves time efficiency and ensures the stability of cloud-based applications compared to manual procedures.
Comparison of TF-IDF and Word2Vec Feature Representations for Emotion Classification of Tokopedia E-Commerce Review Using LinearSVC Azzahra, Fitriyani; Irawan, Bambang; Faqih, Ahmad; Pratama, Denni; Kurnia, Dian Ade
Journal of Artificial Intelligence and Engineering Applications (JAIEA) Vol. 5 No. 2 (2026): February 2026
Publisher : Yayasan Kita Menulis

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59934/jaiea.v5i2.2215

Abstract

This study aims to compare the performance of TF-IDF and Word2Vec feature representations for emotion classification of Tokopedia e-commerce reviews using the LinearSVC algorithm. The dataset used is PRDECT-ID, which consists of 5,400 Indonesian-language reviews labeled with positive and negative emotions. The preprocessing stages include case folding, non-alphabet character cleaning, slang normalization, stopword removal, Sastrawi stemming, and emoji handling. Feature extraction was performed using TF-IDF and Word2Vec, after which the models were trained using LinearSVC and evaluated through 5-Fold Cross Validation and holdout testing. The experimental results show that TF-IDF achieves better performance, with an accuracy of 0.65, a macro-F1 score of 0.645, and a Cohen’s Kappa value of 0.294. Meanwhile, Word2Vec attains an accuracy of 0.58 and a macro-F1 score of 0.540. These findings indicate that TF-IDF is more effective for short and informal texts characteristic of Indonesian e-commerce reviews.