DRDian Palupi Rini

Dian Palupi Rini

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Dian Palupi Rini has a total of 74 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    DWR-CvT with Vanilla GAN-Based Synthetic Data Generation for Cervical Cancer Classification

    by Rudiansyah, Dian Palupi Rini, Lucky Indra Kesuma, Anita Desiani, Deshinta Arrova Dewi, Fitri Salamah, Nadiyah Fadhilah

    Cervical cancer is a major cause of mortality among women worldwide, making early and accurate diagnosis essential. This study proposes Vanilla GAN-based data augmentation combined with a Depthwise Separable Convolution and Residual-based Convolutional Vision Transformer (DWR-CvT) for cervical cell classification. DWR-CvT integrates Depthwise Separable Convolution (DSC) and Residual Connection (RC) before the transformer block to enhance feature extraction while maintaining computational efficiency. The study conducted two experimental setups, including 5-class and 7-class classification tasks, and validated the approach on an external dataset, SIPaKMeD, to evaluate generalization. Cervical cell images were generated using Vanilla GAN, achieving FID 32, LPIPS 0.36, and Cosine Similarity 0.812. The augmented datasets were used to train the proposed DWR-CvT model. Experimental results show 97% accuracy, 96–97% sensitivity, 97–99% specificity, 97–98% precision, 97% F1-score, and 99% AUC across all experiments. The model contains only 0.53 million parameters, indicating greater computational efficiency than conventional deep learning architectures. These findings demonstrate that integrating Vanilla GAN augmentation with the DWR-CvT provides an effective, lightweight, and robust solution for cervical cancer cell classification across multiple class granularities and datasets. The proposed method achieves consistent performance while reducing computational cost, making it suitable for practical clinical and research applications.

  • Open Access

    Allergen detection based on food packaged products for eczema patients using optical character recognition method

    by Dian Palupi Rini, Riska Tri Mardilah, Muhammad Qurhanul Rizqie, Dwi Rosa Indah, Alvi Syahrini Utami, Jovanic Morgan

    Eczema, also known as dermatitis, is a chronic skin condition that causes recurring episodes of dry and itchy skin. It can be managed through medication and by avoiding triggers like stress and certain foods. To help patients avoid food-related triggers, researchers conducted a study to detect allergenic food compositions in packaged products using optical character recognition (OCR) techniques, specifically open computer vision (OpenCV) and Tesseract. The study involved analyzing 120 images of food labels. The process included several steps: preprocessing the images by converting them to a text-friendly format (gray scaling, denoising, and thresholding), using Tesseract for text detection, followed by case folding and tokenization. The results showed that the system achieved an average text detection accuracy of 61.88% and an average allergen detection accuracy of 83.06%. The highest accuracy for text detection was 78.52%, and the highest accuracy for allergen detection was 100%. These findings suggest that OCR techniques can be a useful tool for helping eczema patients manage their diet and minimize flare-ups.

  • Open Access

    Optimization of Cement Distribution Route Based on Hybrid Genetic-Firefly Algorithm (GAFA)

    by Agustina Heryati, Deris Stiawan, Heri Setiawan, Dian Palupi Rini, Rahmat Budiarto

    Optimization of Cement Distribution Route Based on Hybrid Genetic-Firefly Algorithm (GAFA)

  • Open Access

    Residual pixel-wise semantic segmentation for assessing enlarged fetal heart: a preliminary study

    by Muhammad Taufik Roseno, Siti Nurmaini, Dian Palupi Rini, Tommy Saputra, Putri Mirani, Muhammad Naufal Rachmatullah, Annisa Darmawahyuni, Ade Iriani Sapitri, Hadi Syaputra

    The four-chamber view is a crucial scan plane routinely employed in both second-trimester perinatal screening and fetal echocardiographic examinations. Sonographers typically measure biometrics in this plane, such as the cardiothoracic ratio (CTR) and heart axis, to diagnose fetal heart anomalies. However, due to the echocardiographic artifacts, the assessment not only suffers from low efficiency but also inconsistent results depending on the operators’ skills. This study proposes a residual pixel-wise semantic segmentation, which segmented the fetal heart and thoracic contours in a 4-chamber view for assessing an enlarged fetal heart condition. The accuracy of intersection-over-union (IoU) and dice coefficient similarity (DCS) is used for model validation to further regulate the evaluation procedure. We use 1174 US images, comprising about 560 enlarged heart images, and about 614 normal heart images. Out of these data, 248 images are used for unseen data, and the remaining for training/validation processes. The performance of the proposed model, when tested on unseen data, achieved satisfactory results with 97.71% accuracy, 90.36% IoU, and 94.93% DCS. These metrics collectively demonstrate the satisfactory performance of the proposed model compared to existing segmentation models. The outcomes underscore that the proposed model establishes a state-of-the-art standard for enlarged fetal heart detection.

  • Open Access

    Comparison of CNN Architectures for Pre-Cancerous Cervical Lesion Classification Based on Colposopy Images Using IARC and AnnoCerv Datasets

    by Sigit Prasetyo Noprianto, Siti Nurmaini, Dian Palupi Rini

    Cervical cancer represents a significant public health issue affecting women worldwide, and identifying the severity of lesions early on is crucial to selecting the right treatment. This research investigates and compares the effectiveness of various Convolutional Neural Network (CNN) models in classifying colposcopic images according to the severity of cervical lesions. The dataset used was obtained from the International Agency for Research on Cancer (IARC) and AnnoCerv, consisting of 452 colposcopy images categorized into four classes: Normal, CIN 1, CIN 2, and CIN 3. Five CNN architectures were evaluated: MobileNetV2, InceptionV3, Xception, VGG16, and DenseNet121. Experiments were conducted using default hyperparameters: batch size of 32, learning rate of 0.001, and 100 epochs. The results showed that MobileNetV2 achieved the highest accuracy at 67%, followed by DenseNet121 (60%), Xception (60%), InceptionV3 (55%), and VGG16 (42%). Based on these findings, MobileNetV2 is the most optimal model for classifying colposcopy images in this study. However, the study is limited by class imbalance and dataset size, which may affect model generalizability. Future work may explore ensemble learning techniques and larger, more diverse datasets for improved accuracy.


  • Open Access

    Optimization of Backpropagation (BP) Weight Values Using Particle Swarm Optimization (PSO) to Predict KIP Scholarship Recipients

    by Dika Kurnia Nanda, Dian Palupi Rini

    The Indonesia Smart Card (KIP) Lecture program aims to improve the quality of human resources by providing educational assistance to students from underprivileged families. However, the distribution of KIP Lecture in Palembang still faces problems, such as inaccurate targeting and lack of public understanding of this program. The selection process for scholarship recipients is not optimal, causing students who should be prioritized to be overlooked. In addition, decision-making takes a long time due to the many variables that must be considered and the lack of transparency in data processing. This research discusses the Backpropagation (BP) method for predicting KIP College scholarship recipients, which has previously been applied to the classification of educational aid recipients with high accuracies results. However, BP has disadvantages such as minimum local risk and long training time. To overcome this, the Particle Swarm Optimization (PSO) algorithm is used to optimize the weights of the BP artificial neural network. PSO is a simple but effective optimization algorithm to find optimal weights more quickly and accurately. The results of previous studies show that the combination of BP with PSO can improve prediction accuracy compared to using BP alone. Therefore, this research aims to develop a more efficient and targeted prediction model for KIP College scholarship recipients through BP optimization using PSO, so that the selection process can be carried out more quickly and accurately.

  • Open Access

    Aplikasi QR-code untuk sistem daftar hadir: Solusi digitalisasi administrasi di SMA dan SMK

    by Desty Rodiah, Novi Yusliani, Abdiansah, Alvi Syahrini Utami, Kanda Januar Miraswan, Mastura Diana Marieska, Yunita, Dian Palupi Rini

    Kebijakan Merdeka Belajar dari Kemendikbud RI mendorong guru untuk menerapkan pendekatan pengajaran yang fleksibel dan adaptif melalui integrasi teknologi dalam kegiatan pembelajaran. Dalam konteks ini, program pengabdian kepada masyarakat memberikan pelatihan untuk mengembangkan sebuah aplikasi daftar hadir berbasis QR-Code menggunakan Python untuk guru SMA dan SMK. Aplikasi ini dirancang untuk mencatat kehadiran siswa secara cepat, tepat, dan efisien, serta mendukung kemudahan administrasi dan memberikan pengalaman langsung dalam penggunaan teknologi pemrograman. Kegiatan pengabdian ini menerapkan metode Participatory Action Research (PAR), yang meliputi lima tahap: To Know (menggali kebutuhan mitra melalui survei), To Understand (mengevaluasi pelatihan sebelumnya), To Plan (menyusun materi dan instrumen evaluasi), To Act (melaksanakan pelatihan melalui presentasi dan praktikum), dan To Change (melakukan evaluasi). Evaluasi dilakukan dengan pendekatan N-Gain dan skala Likert. Hasil N-Gain sebesar 20,90% menunjukkan efektivitas pelatihan yang kurang meskipun terdapat peningkatan nilai rata-rata sebesar 7,37 poin. Hal ini dipengaruhi oleh latar belakang peserta yang sudah berpengalaman, sehingga materi dan soal perlu dikembangkan lebih lanjut. Di sisi lain, hasil Likert menunjukkan persepsi peserta yang sangat positif. Kendala koneksi internet sempat memengaruhi praktikum, namun narasumber dan mahasiswa aktif membantu peserta yang mengalami hambatan tersebut.

  • Open Access

    Hyperparameter optimization of convolutional neural network using particle swarm optimization for emotion recognition

    by Dian Palupi Rini, Tri Kurnia Sari, Winda Kurnia Sari, Novi Yusliani

    Emotion identification has been widely researched based on facial expressions, voice, and body movements. Several studies on emotion recognition have also been performed using electroencephalography (EEG) signals and the results also show that the technique has a high level of accuracy. EEG signals that detected by standart method using exclusive representations of time and frequency domains presented unefficient results. Some researchers using the convolutional neural network (CNN) method performed EEG signal for emotional recognition and obtained the best results in almost all benchmarks. Although CNN has shown fairly high accuracy, there is still a lot of room for improvement. CNN is sensitive to its hyperparameter value because it has considerable effect on the behavior and efficiency of the CNN architecture. So that the use of optimization algorithms is expected to provide an alternative selection of appropriate hyper parameter values on CNN. Particle swarm optimization (PSO) algorithm is a metaheuristic-based optimization algorithm with many advantages. This PSO algorithm was chosen to optimize the hyperparameter values on CNN. Based on the evaluation results in each model, hybrid CNN-PSO showed better results and achieved the best value in 80:20 split data which is 99.30% accuracy.


  • Open Access

    Klasifikasi Sinyal EEG Untuk Mengenali Jenis Emosi Menggunakan Recurrent Neural Network

    by Aspirani Utari, Dian Palupi Rini, Winda Kurnia Sari, Tommy Saputra

    This research focuses on in-depth exploration and analysis of the application of two types of Recurrent Neural Network (RNN), namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The two models are drilled with the same parameters, consist of three layer, use the relu activation function, and apply 1 dropout level. In order to compare the performance of the two, experiments were carried out using five groups of datasets for training and performance evaluation purposes. The evaluation includes metrics such as accuracy, recall, F1-score, and area under the curve (AUC). The dataset used is Eeg Emotion which contains 2458 unique variables. In terms of performance, LSTM succeeded in outperforming GRU in the task of classifying emotional data based on EEG signals. On the other hand, GRU shows advantages in accelerating the training process compared to LSTM. Although the accuracy of both methods is almost similar in all data divisions, in the evaluation of the ROC curve, the LSTM model demonstrates superiority with a more optimal curve compared to GRU.


  • Comparison of Feature Selection Methods on Medical Record Data

    by Lastri Widya Astuti, Ermatita Ermatita, Dian Palupi Rini

    Medical record data are an essential part of medicine, but the effective use of data has become difficult due to the rapid increase in data. Determining the selected features allows researchers to extract the information from medical data. Feature selection in classification provides the advantage of reduced time and storage. In addition, feature selection methods can improve performance by removing redundant or irrelevant data based on their weights. The benefits of feature selection are that it speeds up the training time, simplifies the model, makes interpretation easier, and improves generalization by reducing overfitting. Feature selection in this research uses two methods: Binary Whale Optimization (BWOA), Particle Swarm Optimization, and Hybrid with Gray Wolf Optimization (PSOGWO). Two feature selection methods were compared to determine the most optimal method for the classification process. A comparison of the two feature selection methods shows that the classification process with many attributes produces an optimal classification process, where the number of representative variables in the breast cancer dataset in the BWOA method has five features, and the PSOGWO method has 11 attributes. Meanwhile, the highest accuracy results in breast cancer classification using the BWOA average of 95.90% and PSOGWO Method average of 97.64%