Course quality is a critical factor shaping learner experience and institutional success. High-quality courses help learners achieve goals and ensure compliance with education standards. Traditional evaluation relies on expert review after course completion, which is slow, costly, and delays improvement. In Massive Open Online Courses (MOOCs), challenges are greater due to sparse and fragmented learning data. We propose LoDiBi (LOQCA, DeepIFSA, BiLSTM), a framework with three integrated modules. LOQCA automatically labels course quality from learner behavior. DeepIFSA imputes missing values using attention, CutMix, and contrastive learning, making it effective in sparse settings. BiLSTM captures temporal learning patterns to enhance prediction accuracy. Combined, these modules enable early prediction of course quality and provide instructional designers with actionable evidence for timely adjustments. Experiments on real-world MOOC datasets show that LoDiBi outperforms existing methods. Data quality was maximum (Completeness and Consistency reached 1). Balanced classification was achieved (MacroF1, Balanced Accuracy greater than 0.9). Strong agreement with ground-truth labels was confirmed (MCC and Kappa greater than 0.9). Predictive performance was also high (Accuracy, Precision, and Recall between 0.93 and 0.94). LoDiBi provides a scalable solution for automated course evaluation, helping institutions make faster, data-driven decisions to improve and adjust course quality.

Massive Open Online Courses (MOOCs) continue to suffer from high dropout rates, raising concerns for both educational quality and institutional efficiency. Predicting learners’ final outcomes is challenging due to three intrinsic data characteristics: extreme sparsity, complex relational structures, and temporal dynamics. We propose GraphGRU-Ed, a two-stage framework that integrates Graph Convolutional Networks (GCNs) with adversarial learning for robust graph-based imputation, and Gated Recurrent Units (GRUs) for sequential prediction. Learners are classified into five performance levels: A (excellent), B (good), C (average), D (low-performing, high dropout risk), and E (dropout). We emphasize label D, which represents learners who remain active but are at imminent risk of dropping out. Early identification of this group is critical for timely interventions to improve course completion. By exploiting graph topology, GraphGRU-Ed imputes missing features while preserving learner relationships, enabling more reliable sequence modeling. Experiments on MOOCCubeX show GraphGRU-Ed achieves F1=0.92 for label D, accuracy=0.95, and AUC=0.99, consistently outperforming traditional baselines (Mean, Median, KNN) and advanced sequential models. These results demonstrate that the framework provides highly reliable predictions (accuracy), excels at correctly identifying at-risk learners (F1 for label D), and maintains strong discriminative ability between dropout-prone and successful learners (AUC). Its strong early-prediction capability enables MOOC platforms to detect at-risk learners and implement effective interventions to reduce dropout rates.

With the emergence of many knowledge-based systems worldwide, there have been more and more applications using different kinds of data and solving significant daily problems. Among that, the issues of missing data in such systems have become more popular, especially in data-driven areas. Other research on the imputation problem has dealt with partial and missing data. This study aims to investigate the imputation techniques for sparse data using the Singular Value Decomposition technique, namely SVDI. We explore the application of the SVDI framework for image classification and text classification tasks that involve sparse data. The experimental results show that the proposed SVDI method improves the speed and accuracy of the imputation process when compared to the PCAI method. We aim to publish our codes related to the SVDI later for the relevant research community.

The field of AI and machine learning is constantly evolving, and as the size of data continues to grow, so does the need for accurate and efficient methods of data processing. However, the data is not always perfect, and missing data is becoming common and occurs more frequently. Therefore, imputation techniques, aside from precision, needed to be scalable. For that reason, we examine the performance of Principle Components Analysis Imputation (PCAI) [ 9 ], an imputation speeding up framework, for logistic regression. Note that the coefficients of a logistic regression model are usually used for interpretation. Therefore, in addition to examining the improvement in the speed of PCAI, we examine how the coefficients of fitted logistic regression models change when using this imputation speeding-up mechanism. To demonstrate the efficiency of the mentioned method, the model’s performance is compared against frequently used imputation methods on three popular datasets: Fashion MNIST, Gene, and Parkinson. And achieves lower time and better accuracy in most experiments.

Object detection in aerial images having many practical applications in real life is becoming popular along with the surging development of deep learning and UAVs (Unmanned Aerial Vehicles). However, adverse weather conditions such as rain, night, and fog might reduce the quality of input images and significantly affect the performance of many perfectly trained detectors (detectors are trained in clear weather conditions). Moreover, object detection in aerial images often seeks high accuracy, while well-known object detection methods use horizontal bounding boxes to represent the object's location. These issues raise the inconsistency between classification and bounding box regression. Understanding the need for practical solutions, we create two experimental Fog datasets based on the original DOTA dataset to provide the in-depth analysis of fog density on multiple well-established oriented object detectors, namely Gliding Vertex, R3Det and ReDet. Furthermore, our training resources achieve promising results, flexibly ensemble to other methods to enhance models' performance and adapt to many adverse weather conditions.

With the help of the rapid development of technology, especially the prevalence of UAVs (unmanned aerial vehicles), object detection in aerial images gains much more attention in computer vision and deep learning.However, traditional methods use horizontal bounding boxes for object representation leading to inconsistency between objects and features.Therefore, many detectors are being built to tackle this problem, and normally they use the conventional approaches of training and testing to achieve the results.Our pipeline proposed to strengthen not only the classification but also localization via independent training processes using convex-hull transformation in data pre-processing phase.We experimented with the well-designed S2ANet, R3Det, ReDet, RoI Transformer and Oriented R-CNN on the well-established oriented object detection dataset DOTA.Then we adopt the best detectors with the well-known classification network EfficientNet to our proposed pipeline and achieve promising results on the oriented object detection DOTA dataset.Moreover, our pipeline can flexibly be adapted to various oriented object detection baselines improving the results in classification via independent extensive training cycles.

In recent years, object detection from space in adverse weather, incredibly foggy, has been challenging. In this study, we conduct an empirical experiment using two de-hazing methods: DW-GAN and Two-Branch, for removing fog, then eval-uate the detection performance of six advanced object detectors belonging to four main categories: two-stage, one-stage, anchor-free and end-to-end in original and de-hazed aerial images to find the best suitable solution for vehicle detection in foggy weather. We use the UIT-DroneFog dataset, a challenging dataset that includes a lot of small, dense objects captured in various altitudes, as the benchmark to evaluate the effectiveness of approaches. After experiments, we observe that each de-hazing method has different impacts on six experimental detectors.

Nowadays, object detection in aerial images in adverse weather, especially in foggy scenes becoming very challenging and incredibly practical. Furthermore, fog and clouds usually appear in the majority of aerial images captured via drones everywhere on Earth, especially in the early morning. Understanding the need for qualified deep learning approaches, we propose a Foggy-DOTA dataset inheriting from the original DOTA dataset and then empirically evaluate it on multiple State-of-the-art methods. After having conducted lots of experiments on some well-known baselines, ReDet is the highest method achieving 76.680 mAP on the original DOTA, only 74.194 mAP on our Foggy-DOTA dataset (60.706 mAP if trained on DOTA). On the other hand, S2ANet and RoI Transformer achieve 74.190, 76,09 mAP on the original DOTA, only yield 71.629 and 73.381 mAP on our Foggy-DOTA dataset (46.503 and 40.297 mAP - significantly low if trained on DOTA), respectively. Our work provides comprehensive statistical evaluation being an essential baseline for future object detection research.

With the rapid development of information and technology, document digitization has become more critical in many research fields by giving enormous amounts of data. However, computers can not handle a lot of information contained inside physical documents. For that reason, making computers detect objects in document images can help humans have more valuable information such as graphs, captions, or tables. There should be a system capable of detecting various components on document images, especially finding a simply effective object recognition method. Thus, the introduction of YOLOF can be an appropriate method to detect objects in documents because it opens up a simple way to exploit image features, making the object detection problem less computationally intensive, but still maintaining the appropriate accuracy. This paper evaluates the new one-stage YOLOF method on two challenging document datasets: IIIT-AR-13K, UIT-DODV. Our experimental YOLOF model achieves 58.8% and 56% on mAP measurement scores with the IIIT-AR-13K dataset and the UIT-DODV dataset, respectively.

In this research, a quick survey was conducted in a Vietnam university and it revealed the outdated paperwork handling in the financial department. Specifically, the bank statement, which is a financial transaction tracking document from banking partners sending monthly to customers, is currently inputted into financial software completely manually. Therefore, this study aims to automatize the extracting data stage by analyzing the table structure on the document, which may reduce the effort for the accountants. The output of this study is extracted text, which can then be added to the software by robotic automation process or other technologies. In this paper, the used methodology is the imaged-based approach only. The bank statement soft copy was converted into an image before being processed through table detection, cell recognition, and text extraction. The text is displayed in a spreadsheet as the output of the process. The measurement on the experiment dataset returned an accuracy of over 93% in most cases. These results suggest that the imaged-based method is applicable for extracting data from the university's bank statements without performing more complicated technologies. However, the output must be put in review by the users to eliminate unwarranted financial errors.

Virtual Reality (VR) and Augmented Reality (AR) are technology trends that have achieved many achievements, attracting investment in research and development in countries around the world. Numerous industries use these technologies mostly for education, entertainment, and advertising purposes.In this article, we study VR/AR technology applied in ecommerce to improve the customer’s product experience. Customers can “touch” the product through virtual product simulation models, get more visual information about the product (instead of just viewing the product through images on the website or leaflets, posters) thus encouraging customers to make a purchase decision. We hope that this research will help customers have more positive experiences when choosing to shop for products. We expect that improving the product experience results in improvements to the perception people have of the brand.

This paper is concerned with the development of a new hybrid metaheuristic approach for solving a practical university course timetabling problem in Vietnam. Our hybrid method is a combination of Harmony Search (HS) algorithm and the Bees algorithm. The proposed method has been tested on 14 real-world data instances and compared with some other metaheurisitic approaches, which are Variable Neighborhood Search, Tabu Search and Bee Algorithms. Numerical results indicate the effectiveness of the hybrid HS – Bees algorithm over the others for the specific problem.

In this paper, Particle Swarm Optimization algorithm and its combination with hill-climbing and Tabu Search algorithm are applied to a real-world university course timetabling problem. Experimental results taken from the tests on 14 practical data instances are shown, leading to the drawing of some conclusions.

This paper presents the application of Variable Neighborhood Search algorithm and seven of its variants on a real-world highly constrained curriculum-based university course timetabling problem. Experimental results on 14 instances taken from practice are shown and some conclusions about the efficiency of these algorithms on the considered problem are drawn.