Research Publications

Authors

Lahuddin, Harlinda, Program Studi Sistem Informasi, UniversitasMuslim Indonesia, Makassar, Indonesia.; Astuti, Wistiani, Program Studi Teknik Informatika, Universitas Muslim Indonesia, Makassar, Indonesia.

Abstract

Program Pelatihan Computational Thinking dan Problem Solving untuk Generasi Muda Kedah dilaksanakan sebagai respons terhadap rendahnya literasi digital di daerah rural Malaysia, khususnya di Sekolah Kebangsaan Ulu Sedaka, Yan, Kedah. Tujuan program adalah membekali siswa dengan keterampilan berpikir komputasi yang mencakup decomposition, pattern recognition, abstraction, dan algorithmic thinking sebagai fondasi menghadapi era digital. Metode pelaksanaan meliputi baseline assessment, workshop interaktif, hands-on activities, serta evaluasi melalui pre-test dan post-test. Sebanyak 30 siswa dan 5 guru terlibat aktif dalam program selama 12 minggu. Hasil evaluasi menunjukkan peningkatan signifikan pemahaman computational thinking dari skor rata-rata 45,2 menjadi 78,6 dengan peningkatan 73,9 persen. Tingkat partisipasi mencapai 93,3 persen dengan kepuasan peserta 4,5 dari skala 5. Program berhasil mengembangkan modul pembelajaran digital dan panduan aplikasi yang dapat digunakan berkelanjutan. Kesimpulan menunjukkan bahwa pendekatan pembelajaran interaktif berbasis unplugged activities efektif meningkatkan kemampuan berpikir logis siswa di daerah dengan keterbatasan infrastruktur teknologi, serta memberikan dampak positif terhadap minat siswa pada bidang STEM dan kesiapan menghadapi transformasi digital.

Related SDGs

Authors

Wulandari, LMC; Nurhayati, Lilis; Dimas, Ravael; Pijoh, Fidelchristo, Prodi Teknik Industri, Fakultas Teknik, Universitas Katolik Darma Cendika, Indonesia, 61557

Abstract

Sector Logistics plays an important role in maintaining the smoothness and efficiency of national and global supply chains, especially for third-party logistics ( 3PL ) service providers who face demands for reliability and speed of service amidst global competition. The complexity of the logistics process creates the need for a structured and measurable performance evaluation model. This study aims to apply the Supply Chain Operations Reference ( SCOR ) model in identifying and prioritizing key performance indicators ( KPI ) in a 3PL company . The method used involves distributing questionnaires to decision makers in operational and managerial fields. The weights of SCOR performance attributes including reliability, responsiveness, flexibility, cost measures, and asset management efficiency are calculated using the Fuzzy Analytic Hierarchy Process ( FAHP ) method. Furthermore, the Technique for Order Preference by Similarity to Ideal Solution ( TOPSIS ) is used to determine KPI priorities at PT.X. The results of the study show that there are 17 indicators to measure the performance of the freight forwarding sector with the largest weight being operator reliability (0.251) on the reliability attribute, the number of on-time deliveries (0.694) on the responsiveness attribute, load flexibility (0.317) on the flexibility attribute, shipping costs per km (0.379) on the cost attribute and cash-to-cash cycle time (0.479) on the asset management attribute. The SCOR model has proven effective as an initial framework in measuring the performance of 3PL logistics service providers, because it is able to integrate various aspects of performance systematically and quantitatively.

Related SDGs

Authors

Mubarak, Mush’ab Al; Syahar, A. Ulfah Tenripada; Kasim, Fadly, Program StudiTeknik Informatika, FakultasIlmu Komputer, UniversitasMuslim Indonesia, Jl.Urip Sumoharjo Km.05, Kota Makassar, Sulawesi Selatan, Indonesia; Hayati, Lilis Nur; Mansyur, St Hajrah, Program StudiSistem Informasi, FakultasIlmu Komputer, UniversitasMuslim Indonesia, Jl.Urip Sumoharjo Km.05, Kota Makassar, Sulawesi Selatan, Indonesia

Abstract

Vanilla is a high-value plantation commodity whose productivity is significantly affected by plant diseases that are difficult to identify accurately using conventional methods. This study aims to develop a mobile-based vanilla plant disease identification system using a Convolutional Neural Network (CNN) with the ResNet152 architecture. The dataset consists of primary field-acquired images, which were augmented to produce a total of 1,616 images across five disease classes. The model was trained using a transfer learning scheme with parameter adjustments designed to handle variations in field lighting conditions, image angles, and real-world visual characteristics. Experimental results demonstrate that the proposed ResNet152 model achieves high and stable classification accuracy. The integration of the trained model into a mobile application enables fast and practical disease diagnosis in real plantation environments. The novelty of this study lies in the field-oriented optimization of the ResNet152 model and its direct deployment in a mobile diagnostic system tailored for vanilla plant disease identification.

Related SDGs

Authors

Salsabilah, Nadya; Irawati; Hayati, Lilis Nur, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

Tomatoes are one of the most widely cultivated and consumed crops, but they are highly susceptible to disease attacks. The main diseases that often attack tomato plants are early blight and late blight. This study compares two machine learning-based classification methods, namely Convolutional Neural Network (CNN) and Naïve Bayes, in detecting tomato leaf diseases. The dataset used consists of 1,255 images obtained from Kaggle, which have been processed and divided into three data ratio scenarios (70:30, 80:20, and 90:10) for training and testing. The results showed that CNN is superior to Naïve Bayes, with the highest accuracy reaching 83.01%, while Naïve Bayes only achieved 34%. With better stability and accuracy, CNN has the potential to help farmers detect diseases more quickly and increase agricultural productivity

Related SDGs

Authors

Azis, Huzain; M, Muh Fatwah Fajriansyah; Darwis, Herdianti; Purnawansyah; Hasanuddin, Tasrif; Sugiarti, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

Automatic Speech Recognition (ASR) systems perform well for high-resource languages but remain unreliable for regional dialects with limited data. The Makassar dialect, spoken by approximately nine million people in South Sulawesi, Indonesia, presents unique phonetic and grammatical challenges, including high-frequency particles such as mi, ji, and ko. This study introduces the first benchmark evaluation of state-of-the-art ASR models on Makassar speech. Four models, Whisper (tiny, base, small) and Wav2Vec2 Large XLSR Indonesian, were tested on 305 spontaneous utterances (∼10 minutes) from 10 native speakers. Results show severe performance degradation: the best model (Wav2Vec2 Indonesian) reached 87.73% of word error rate (12.27% accuracy). Error analysis reveals two dominant failure modes: Dialect Particle Blindness (average detection rate 2.9%) and Systematic Phonetic Mismatch (89 vowel confusions), indicating that current models treat dialectal features as noise. These findings underscore the urgent need for dialect-aware ASR adaptation and dataset development, providing a foundation for inclusive speech technology across Indonesia's linguistic diversity.

Related SDGs

Authors

Darwis, Herdianti; Adnan, Adam; Purnawansyah; Manga, Abdul Rachman, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia; Wibawa, Aji Prasetya; Irawati, Department of Electrical and Informatics Engineering, Universitas Negeri Malang, Malang, Indonesia

Abstract

The proliferation of online learning platforms has necessitated automated systems for monitoring students' emotional states, given that variations in facial expressions significantly influence engagement and learning outcomes. This study proposes a spatio-temporal classification framework for recognizing emotion intensity levels in the DAiSEE dataset, utilizing Vision Transformer as a spatial feature extractor alongside various temporal models, including LSTM, BiLSTM, TimeSformer, and their hybrid variants. Embeddings are extracted via ViT, after which temporal dependencies are captured by each classifier, incorporating feature-level oversampling to mitigate severe class imbalance. Experimental findings reveal that, despite ViT's ability to generate robust spatial representations, all temporal models struggle to identify minority classes, resulting in predictions biased toward the majority class as evidenced by low balanced accuracy scores and overlapping clusters in t-SNE visualizations. Among all configurations, the ViT + LSTM model delivered the most reliable performance, attaining 59 % accuracy and a 0.59 weighted F1-score on engagement labels, while remaining competitive with prior methods. In essence, integrating spatial and temporal features enhances classification efficacy, yet its effectiveness is substantially constrained by imbalanced data distributions. These results offer a thorough examination of representational challenges in imbalanced affective datasets, along with recommendations for mitigation techniques, crossdataset assessments, and multimodal integrations.

Related SDGs

Authors

Manga, Abdul Rachman; Nirmala; Azis, Huzain; Fattah, Farniwati; Salim, Yulita; Darwis, Herdianti, Departement of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

Flower image classification poses a challenge in digital image processing, requiring effective methods for feature extraction and classification. The aim of this research is to improve the accuracy of flower image classification by employing ResNet-50 for feature extraction, with and without segmentation, and evaluating the effectiveness of various classification algorithms. The dataset consists of images of Calendula and Coreopsis flowers, totaling 2,025 samples. Four segmentation techniques-Canny, Thresholding, Otsu, and Mean Shift-along with a non-segmentation approach are applied. The features extracted using ResNet-50 are classified with Naive Bayes, Support Vector Machine (SVM), Decision Tree, and K-Neighbors Classifier (KNN). Performance evaluation is conducted using accuracy, precision, recall, and F1-score with 5-fold cross-validation. The results show that SVM delivers the best performance in most scenarios. The highest accuracy in segmentation scenarios was achieved with the Mean Shift technique at 0.91, while the non-segmentation approach yielded the highest accuracy of 0.95. The non-segmentation approach proves more effective, indicating segmentation is not always required for high classification accuracy with ResNet-50. This study shows that ResNet-50, especially without segmentation, can significantly improve flower image classification compared to traditional methods, opening opportunities for more efficient systems.

Related SDGs

Authors

Salim, Yulita; Rakasyah, Athar Fathana; Darwis, Herdianti; Herlinda; Irawati; Manga, Abdul Rachman, Departement of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

In the era of burgeoning image data, the demand for robust image classification algorithms has never been more pressing. This research article delves into the realm of image classification, encompassing a comprehensive analysis of diverse datasets and machine learning algorithms, While simultaneously studying the efficiency of ensemble approaches in improving classification performance. The study employs five distinct datasets, spanning medical images, everyday objects, and natural scenery, ensuring a broad spectrum of classification challenges. These datasets include the Skin Cancer ISIC dataset, CIFAR-10, Flowers, Apparel Image, and Brain Tumor dataset. The image classes within these data sets vary significantly, presenting an ideal testbed for assessing algorithmic versatility. Our investigation scrutinizes five machine learning algorithms: Support Vector Classifier (SVC), Random Forest Classifier (RFC), Gradient Boosting Classifier (GBC), K-Nearest Neighbors (KNN), and Gaussian Naive Bayes (GNB). Each algorithm is evaluated individually, followed by a comprehensive ensemble approach involving a Voting Classifier. The results unveil nuanced performance variations across diverse datasets. Notably, RFC and GBC exhibit remarkable accuracy in brain tumor image classification, while KNN demonstrates strengths in classifying apparel images. Ensemble techniques, embodied by the Voting Classifier, harmonize these algorithms, yielding competitive and balanced performance across the datasets. This article contributes valuable insights into the realm of image classification, shedding light on algorithmic strengths and limitations, the efficacy of ensemble techniques, and their applicability to diverse image datasets. These findings hold significance for fields ranging from medical diagnostics to everyday object recognition, paving the way for more precise and versatile image classification solution

Related SDGs

Authors

Darwis, Herdianti; Puspitasari, Rahma; Purnawansyah; Astuti, Wistiani; Atmajaya, Dedy; Hasnawi, Mardiyyah, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

This research investigates the use of deep learning for automatic waste classification, specifically using ResNet50 for feature extraction and combining it with various classification algorithms. The dataset comprises 1889 images categorized into four classes: plastic, metal, cardboard, and paper. Two approaches were evaluated: direct classification and feature extraction with ResNet50. The direct classification models, including SVM, KNN, and Random Forest, resulted in low performance, with an average accuracy of 60%. However, using ResNet50 for feature extraction significantly improved the classification accuracy across all models, with the combination of ResNet50 and SVM achieving an accuracy of 91%, and precision, recall, and F1-Score exceeding 92%. This demonstrates the effectiveness of ResNet50's feature extraction capability in enhancing the classification of images. The findings suggest that combining feature extraction and classification models provides a more accurate and efficient solution for automatic waste management systems, supporting the recycling process and waste management efficiency.

Related SDGs

Authors

Purnawansyah; Supriadi, Naufal Abiyyu; Manga, Abdul Rachman; Adawiyah, Rabiatul; Harlinda; Hasanuddin, Tasrif, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

The volume of network traffic is ever-increasing, hence making DDoS attacks serious threats to the integrity of network security. This research paper focuses on improving the accuracy of detecting DDoS attacks using the ensemble machine learning methods of stacking, bagging, voting, and gradient boosting. In this work, is dataset big enough to correctly reflect real network traffic is used for training and testing purposes. The results indicate that gradient boosting has an accuracy rate of 0.99, followed by stacking and bagging, at 0.98. In this regard, both approaches tend to be efficient in the identification of the attack with limited errors in the prediction. Accordingly, the ensemble technique thus provides a robust approach toward the detection of DDoS attacks, while the study provides important information on the implementation of such strategies in real-time network security systems. Additional investigation is advised to examine supplementary datasets and optimize hyperparameters in order to enhance detection efficacy.

Related SDGs

Authors

Satra, Ramdan; Dahlan, Imram Afdillah; Darwis, Herdianti; Purnawansyah; Mujaddid, Syariful; Fattah, Farniwati, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

This study focuses on comparing the performance of K-Nearest Neighbors (KNN), TabNet, and Wide & deep learning methods in classifying Distributed Denial of Service (DDoS) attacks on Software-Defined Networks (SDN). The use of SDN enables centralized control of network infrastructure, making it vulnerable to DDoS attacks occur due to the centralized nature of SDN architecture. Machine learning models, including KNN, TabNet, and Wide & deep learning, are applied to an SDN-specific DDoS dataset to evaluate their effectiveness in accurately classifying normal and malicious traffic. These models were tested using various data splits (60:40, 70:30, 80:20, and 90:10) to determine the optimal ratio for training and validation. KNN exhibited the highest accuracy, reaching 98% in both 80:20 and 90:10 splits, while wide & deep learning achieved 94.99% accuracy, and TabNet demonstrated a 93.59% accuracy. The results suggest that KNN, despite being a simpler algorithm, outperforms the more complex deep learning models in this specific task. The findings provide valuable insights for researchers and network administrators in selecting effective machine learning algorithms for DDoS detection in SDN environments.

Related SDGs

Authors

Darwis, Herdianti; Pagala, Arya Nanda Pratama; Anraeni, Siska; Amaliah, Tazkirah; As'ad, Ihwana; Tenripada, Andi Ulfah, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

Childfree is the choice to live without having or adopting children. This phenomenon is still considered a sensitive topic in Indonesia, where the prevailing belief holds that the primary goal of marriage is to have children. Numerous individuals share their perspectives on this matter through Twitter. In this research, 6654 raw data have been collected from Twitter using crawling techniques using Rapidminer and scraping using website scrape based on the keyword “childfree” which is preprocessed into clean data. The sentiment analysis model carried out includes categorizing childfree sentiment based on religious, medical and economic fields, then measuring the performance of the support vector machine, involving several methods such as RoBERTa labeling, bigram tokenizing, term frequency inverse document frequency weighting, 5 cross validation training, and synthetic minority over-sampling technique with edited nearest neighbors. The research results show that the economic category is the most influential field with 722 related sentiments, the accuracy performance of SVM gives a value of 75.95% on the linear kernel and the application of SMOTEENN gives a value of 95.94% on the linear kernel, it is proven that using SMOOTEENN can overcome data imbalance.

Related SDGs

Authors

Gaffar, Andi Widya Mufila; Azmi, Nurul; Alwi, Erick Irawadi; Abdullah, Syahrul Mubarak; Adawiyah, Rabiatul; Widyawati, Dewi, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

MRI is a very important diagnostic tool in finding brain tumors, but recently the development of manual interpretation of MRI images has developed some challenges, such as diagnosis is often delayed, and there is a high chance of making mistakes. Recently, to hurry up the process, brain tumor detection has started applying machine learning methods. The classification of MRI images for brain tumors is done in this research paper by extracting their features with the ResNet-50 model. The classical machine learning algorithms that have been applied in the paper for classifying the tumors using the extracted features include Naive Bayes, Decision Tree, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest. In this respect, SVM and KNN have yielded the highest accuracy of 0.95 and 0.96 respectively, hence are the best methods in this task. The study's findings contribute to the development of quicker and more precise methods to help with brain tumor diagnosis in a medical context.

Related SDGs

Authors

Manga, Abdul Rachman; Nanda, As'syahrin, Department of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia; Handayani, Anik Nur; Herwanto, Heru Wahyu, Department of Electrical Engineering, Universitas Negeri Malang, Malang, Indonesia; Asmara, Rosa Andrie, Information Technology Department, State Polytechnic of Malang, Malang, Indonesia; Lantara, Dirgahayu, Faculty of Industrial Engineering, Universitas Muslim Indonesia, Makassar, Indonesia

Abstract

Data classification plays a crucial role in artificial intelligence, particularly in enhancing model accuracy. This study focuses on classifying Toraja buffalo, a livestock breed with significant cultural importance in South Sulawesi, Indonesia. While the Single Input Approach is commonly used for classification, it often fails to capture all the necessary attributes to effectively distinguish between racial traits. Therefore, this research aims to evaluate the effectiveness of a multi-input approach, which integrates multiple data inputs to improve classification performance compared to the Single Input method. We employed four classification techniques: K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Naïve Bayes, and Decision Tree, using both Single Input and Multi Input configurations. Model performance was assessed through Precision, Recall, F1 Score, and Accuracy metrics. The findings indicate that the Multi Input approach consistently outperformed the Single Input method. Notably, KNN achieved its best performance with Multi Input, recording an F1 Score of 0.7263 and an Accuracy of 0.7333, significantly surpassing the results obtained from Single Input. Similarly, SVM also demonstrated substantial performance enhancements with Multi Input. Overall, the study highlights the importance of incorporating a wider array of informative data to enhance the model's capability in accurately classifying specific categories, with KNN showing the most pronounced improvements

Related SDGs

Authors

Purnawansyah; Adnan, Adam; Darwis, Herdianti, Faculty of Computer Science, Universitas Muslim Indonesia, Jl. Urip Sumoharjo KM 5, Makassar, 90231, Indonesia; Wibawa, Aji Prasetya, Universitas Negeri Malang, Jl. Semarang No. 5, Malang, 65145, Indonesia; Widyaningtyas, Triyanna; Haviluddin, Universitas Mulawarman, Jl. Kuaro, Samarinda, 75119, Indonesia

Abstract

Facial Expression Recognition (FER) plays a crucial role in humancomputer interaction, yet improving its accuracy remains a significant challenge. This study aims to enhance the robustness and effectiveness of FER systems by integrating multiple machine learning techniques within a semi-supervised learning framework. The primary objective is to develop a more effective ensemble model that combines Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Support Vector Classifier (SVC), and Random Forest classifiers, utilizing both labeled and unlabeled data. The research implements data augmentation and feature extraction techniques, utilizing advanced architectures such as VGG19, ResNet50, and InceptionV3 to improve the quality and representation of facial expression data. Evaluations were conducted across three dataset scenarios: original, feature-extracted, and augmented, using various labelto- unlabeled ratios. The results indicate that the ensemble model achieved a notable accuracy improvement of 87% on the augmented dataset compared to individual classifiers and other ensemble methods, demonstrating superior performance in handling occlusions and diverse data conditions. However, several limitations exist. The study's reliance on the JAFFE dataset may restrict its generalizability, as it may not cover the full range of facial expressions encountered in real-world scenarios. Additionally, the effect of label-to-unlabeled ratios on the model's performance requires further exploration. Computational efficiency and training time were also not evaluated, which are critical considerations for practical implementation. For future research, it is recommended to employ cross-validation methods for more robust performance evaluation, explore additional data augmentation techniques, optimize ensemble configurations, and address the computational efficiency of the model to better advance FER technologies.

Related SDGs

Authors

Herman; Putra Muhammad Dani Arya, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia; Nasir, Haidawati, MIIT University Kuala Lumpur, Malaysia; Darwis, Herdianti; Mansyur, St. Hajrah, Faculty of Computer Science, Universitas Muslim Indonesia, Makassar, Indonesia; Noor, Megat Nurolazmi MM, MIIT University Kuala Lumpur, Malaysia

Abstract

This study focuses on being able to classify traditional Toraja carvings using Convolutional Neural Networks (CNN), focusing on three CNN architectures, specifically VGG-16, VGG-19, and a model that is concatenated from both. The aim is to determine the most effective architecture and ratio of training data and validation data sharing to achieve the highest classification accuracy. The image dataset consisting of seven different Toraja carving motifs or classes underwent data pre-processing, namely data augmentation, to improve model generalization and reduce overfitting. Experiments were conducted using four scenarios of training data and validation data separation. The final outcome of this research is that VGG-16 reached the best validation performance of 97.36% with a data 90%: 10% separation. It manifests its superior ability to Capture the information of complicated Toraja carving motifs. VGG-19 and the combined model also performed well, but the results were still below the best results of VGG-16 and emphasized that the VGG-16 architecture, especially with a data separation of 90%:10%, is the most reliable CNN architecture for accurately classifying Toraja carvings.

Related SDGs