Rosmawinda Rosmawinda

Rosmawinda Rosmawinda

Auto-generated profile, based on publicly available information

Rosmawinda Rosmawinda

Department of Electrical Engineering and Computer, Universitas Syiah Kuala, Banda Aceh, Indonesia, Indonesia

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
31 August 2026 in JURNAL INFOTEL
Verified
Journal Article
Open access
YYYunidar Yunidar
Yunidar Yunidar
MRMuhammad Mirza Rahmad
Muhammad Mirza Rahmad
FFFithria Fithria
Fithria Fithria
MMMelinda MELINDA
Melinda MELINDA
SRSiti Rusdiana
Siti Rusdiana
RRRosmawinda Rosmawinda
Rosmawinda Rosmawinda
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that requires objective and non-invasive methods for early identification. This study proposes an EEG-based ASD classification framework utilizing a Time-Frequency Synergy Network (TFSNet) at the feature level, integrated with multiscale denoising methods. EEG data were collected from 16 children (8 with ASD and 8 with typical development as a control group) using a 16-channel OpenBCI system. The signals were filtered using a 0.5–40 Hz Butterworth bandpass filter and subsequently denoised via Multiscale Principal Component Analysis (MSPCA) and Multiscale Independent Component Analysis (MSICA). The signals were segmented into 2-second epochs with a 50% overlap. Power Spectral Density (PSD) and bandpower features were processed using a dual-branch TFSNet, which incorporates convolutional feature extraction, attention-based refinement, Transformer encoding, and Adaptive Cross-Domain Attention (ACDA). Performance was evaluated using subject-based 8-fold cross-validation. The TFSNet with MSPCA achieved an average epoch-level accuracy of 98.60%, precision of 98.78%, recall of 98.41%, F1-score of 98.59%, specificity of 98.79%, and an AUC of 99.85%. Although MSICA yielded a higher signal-to-noise ratio (24.48 dB compared to 22.24 dB), its classification accuracy was lower (98.07%). Statistical tests indicate a significant performance difference, with the MSPCA method performing better (paired t-test, p = 0.0073; Wilcoxon test, p = 0.0078).

4 February 2026 in Journal of Electronics, Electromedical Engineering, and Medical Informatics
Verified
Journal Article
Open access
YYYunidar Yunidar
Yunidar Yunidar
IHInda Mariana Harahap
Inda Mariana Harahap
MMMelinda MELINDA
Melinda MELINDA
RRRosmawinda Rosmawinda
Rosmawinda Rosmawinda
NBNurlida Basir
Nurlida Basir
IRImam Fathur Rahman
Imam Fathur Rahman
Down syndrome (trisomy 21) is a genetic disorder caused by an extra copy of chromosome 21, resulting in distinctive developmental facial characteristics and intellectual delays. Early detection is crucial to enable timely medical intervention. However, conventional diagnostic procedures still rely on clinical observation and genetic testing, which can be invasive and expensive. This study proposes a facial image–based classification system for detecting Down syndrome using a Convolutional Neural Network (CNN) approach. Seven CNN architectures were evaluated, namely EfficientNetB0, MobileNetV2, ResNet34, ShuffleNetV2, AlexNet, VGG19, and InceptionV3, under two training scenarios: with and without early stopping. The dataset consisted of 1,000 facial images of children with and without Down syndrome, split into training, validation, and test sets at 60:20:20. Face detection was performed using the Haar Cascade Classifier, followed by data augmentation techniques including rotation, zoom, translation, horizontal flipping, and Gaussian noise to improve model generalization and reduce overfitting. Experimental results show that the VGG19 architecture achieved the best performance, with an accuracy of 94.5%, precision of 91.59%, recall of 98%, and an F1-score of 94.69%. A one-way ANOVA test yielded an F-value of 0.003 and a p-value of 0.955 (> 0.05), indicating no statistically significant difference between models trained with and without early stopping. Grad-CAM visualization highlighted key facial regions, namely the eyes, nose, and mouth, as the primary contributors to classification, while analysis using 68 facial landmark points revealed distinctive morphological patterns associated with Down syndrome. The integration of CNN models, Grad-CAM visualization, and facial landmark analysis demonstrates a promising, interpretable, and non-invasive approach to supporting early Down syndrome screening using facial images

1 June 2026 in Indonesian Journal of electronics, electromedical engineering, and medical informatics
Verified
Journal Article
Open access
PDF file
DUDzaki Dhiya Ul-Haq
Dzaki Dhiya Ul-Haq
YYYunidar Yunidar
Yunidar Yunidar
MMMelinda MELINDA
Melinda MELINDA
NBNurlida Basir
Nurlida Basir
RRRosmawinda Rosmawinda
Rosmawinda Rosmawinda
Early identification of Down Syndrome (DS) is essential for timely intervention; however, conventional diagnostic approaches often require specialized clinical expertise and significant resources. Recent advances in deep learning-based facial image analysis offer a promising alternative, yet the impact of data partitioning strategies on model performance and stability remains insufficiently explored. This study investigates the effects of different data-splitting strategies on DS facial image classification using EfficientNet-B0. A total of 3,030 facial images were collected from Roboflow and curated through preprocessing techniques, including Gaussian noise reduction, image sharpening, and contrast enhancement. Two data partitioning configurations, 70:20:10 and 80:10:10, were evaluated using five-fold cross-validation. Model performance was assessed using accuracy, precision, recall, and F1-score, while statistical significance was examined using the Friedman test. The results show that the 70:20:10 configuration achieved an average accuracy of 87.88% ± 3.03%, while the 80:10:10 configuration achieved a slightly higher accuracy of 89.09% ± 2.53%. The Friedman test indicates statistically significant differences (p < 0.05). However, the improvement is relatively marginal, with a small-to-moderate effect size (Cohen’s d = 0.43) and no significant difference in variance (p > 0.05), indicating limited practical significance. A trade-off between accuracy and evaluation stability was observed. While the 80:10:10 configuration benefits from a larger training set, the 70:20:10 configuration provides more stable and balanced performance, particularly in minimizing false negatives. These findings highlight that higher accuracy does not necessarily imply more reliable or clinically meaningful performance, emphasizing the importance of appropriate data partitioning in medical image classification.