Journal of ICT Research and Applications

A Vision-Transformer-based Ensemble Model for Multi-label Disease Classification on CT Scans and Chest Radiographs

2026-05-25T08:46:31+07:00

The increase in the volume of X-rays and computed tomography (CT) images has drastically increased the workload on radiologists. As a result, a computer-aided solution with the capability to classify CT scans is needed to reduce this workload. In this paper, a vision-transformer (ViT) based model for multi-disease, ensemble-based classification is proposed. ViT, Data2Vec, and SegFormer models were fine-tuned to carry out the classification of selected diseases, namely, effusion, pneumonia, and pneumothorax. Normal cases of the selected diseases were included in the dataset. The datasets were obtained from two sources: the chest X-ray dataset from the Nigerian Institute of Health Chest Clinic and the optical coherence tomography (OCT) images dataset containing 6,621 images from University of California, San Diego. The images were preprocessed using random cropping, horizontal flipping, and normalization. The dataset was partitioned into training, validation, and testing sets. Model training was done in 10 epochs. The evaluation metrics showed a better performance from ensemble learning compared to other individual transformer models. The weighted average performance for all metrics was 90.56% precision, 90.58% recall, and 90.48% F1 score. The model is useful for classification of multiple diseases and can be used by radiologists.

Cross-Lingual Transfer for Semantic Role Labeling in Indonesian

2026-06-15T08:21:39+07:00

Semantic role labeling is a semantic analysis task that aims to identify the semantic relationships within a sentence, such as who did what to whom, where, when, and so on. Current semantic role labeling (SRL) models for the Indonesian language still face challenges in achieving strong performance due to the limited availability of annotated corpora, especially compared with English SRL models. Therefore, this paper develops an Indonesian SRL model using cross-lingual transfer. This approach addresses the data scarcity problem in Indonesian SRL by leveraging the availability of annotated English-language corpora. The method uses multilingual models and SRL datasets from both English and Indonesian. The multilingual models used in this study are XLM-R and mT5, both in base and large configurations. The datasets include Universal PropBank Indonesia and Gojali’s dataset for Indonesian, and CoNLL-2012 for English. Evaluation was conducted using test data from Universal PropBank Indonesia and Gojali’s dataset. Among all developed models, XLM-R large with cross-lingual transfer achieved the best performance, with an F1 score of 0.916 on Gojali’s dataset and 0.858 on the combined Universal PropBank Indonesia and Gojali datasets.

Task-Aware Load Balancing in Mobile Cloud Computing using Cloudlets

2026-06-18T11:21:22+07:00

In the current era, almost every person uses a mobile device for tasks that range from basic phone calls to advanced computations. To increase the QoE of the mobile user, tasks may be offloaded to a cloudlet because of the lack of a large battery, large memory, and sufficient processing power. Cloudlets can be heavily loaded with tasks when they are in a heavily populated area, where tasks are overflowing. Meanwhile, other cloudlets are lightly loaded and their resources are often idle when they are in a moderately populated area, getting fewer tasks. When tasks are queued up for a long time in a heavily loaded cloudlet, the response time and dropout rate of tasks increase. These two parameters deteriorate further when a big task is waiting for a long time. Such big tasks waiting for a long time can be migrated to other cloudlets to utilize the idle resources available in lightly loaded cloudlets, which may provide better QoE to the user. This article uses the Firefly optimization algorithm for choosing which tasks to migrate to which cloudlet for load balancing based on task details such as total waiting task size, task waiting time, and total virtual machine size. The proposed method, called TALBMCC (Task-Aware Load Balancing in Mobile Cloud Computing), also increases the number of tasks executed by cloudlets and reduces the execution time of tasks and the power consumption of mobile devices.

Determination of Oil Palm Planting Patterns using ArduSimple RTK2B GNSS, LoRa Communication, and Drone-based Orthophotos

2026-06-12T09:36:48+07:00

Oil palm is a major plantation commodity in Indonesia, where accurate planting patterns are essential to optimize yield and support sustainable management. Conventional methods using stakes and ropes are labor-intensive and prone to spatial error, whereas many GNSS-based systems are costly and proprietary. This study develops a low-cost real-time kinematic (RTK) GNSS system for determining oil palm planting patterns, based on the ArduSimple RTK2B module, a multiband GNSS antenna, and long-range LoRa communication integrated with a mobile application and orthophoto-derived digital maps. In the base–rover configuration, RTCM corrections are transmitted via LoRa, while planting points designed in ArcGIS are imported into SW Maps for field navigation. Performance is assessed through benchmark accuracy tests, LoRa packet delivery ratio and packet loss measurements over 10–100 m, planting-pattern trials for square and triangular layouts, and a comparison with an actual oil palm layout in a commercial estate. The rover achieved a horizontal positioning error of 8.95 cm, while planting trials yielded mean errors of 8.40 cm (RMSE 9.3 cm) for the square layout and 7.78 cm (RMSE 9.1 cm) for the triangular layout, both within the 10 cm tolerance. The proposed system provides practical, low-cost accuracy for oil palm planting guidance.

Hybrid Genetic Algorithm and Particle Swarm Optimization as Centroid Initialization in K-Means Clustering for National Nutrition Dataset

2026-06-23T08:29:35+07:00

The efficacy of large-scale public health interventions, such as Indonesia’s Makan Bergizi Gratis (MBG) program, relies heavily on the precise identification of vulnerable regions. Conventional clustering techniques like K-means are essential for such mapping but frequently suffer from performance degradation due to sensitivity to random centroid initialization, leading to suboptimal and unstable solutions. This study proposes a sequential hybrid metaheuristic framework (K-means+GAPSO) that synergizes the broad global exploration of genetic algorithms (GA) with the rapid local exploitation of particle swarm optimization (PSO). The proposed model was rigorously validated using the Indonesian National Nutrition Dataset (2021–2023) through 10 independent runs to ensure stochastic robustness. Computational results at the optimal cluster count (K = 3) demonstrated that the K-means+GAPSO (prob = 0.3) configuration significantly outperformed the standard baselines, achieving a highly stable silhouette mean of 0.7130 (±0.0036) and a Davies-Bouldin index mean of 0.4708 (±0.0178). This metric achievement represents a substantial performance improvement in separation and compactness compared to standard K-means. The implementation of this structurally robust method for nutritional clustering provides a reliable analytical foundation for policymakers to accurately target food security initiatives and minimize resource misallocation.

JATO: Deep Reinforcement Learning-based Joint Optimization for Task Offloading and Adaptive Transmission in Multimedia IoT Systems

2026-07-15T08:45:25+07:00

As the Multimedia Internet of Things (M-IoT) evolves, the orchestration of numerous resources that offer support for high-bandwidth, low-latency applications arises as a key challenge. Architecturally, the edge-cloud framework alleviates structural concerns, but the linked nature of compute and data transfer poses problems of resource management. Approaches that tackle task offloading and adaptive transmission that think independently of each other tend to have problems such as user-server cross-region overloads or network congestion. This paper presents JATO, a framework to jointly tackle the problems of adaptive task offloading and transmission optimization using Deep Reinforcement Learning. JATO offers a mono-faceted solution, learning a policy to simultaneously determine the best offloading target and the transmission quality. The framework was implemented for evaluation with a combination of different edge devices in a testbed alongside a simulation environment. JATO recorded a result of 0.9321 as the holistic score of the overall framework endpoint, a score significantly better than that of all the other frameworks that were used as functional baselines. JATO was able to resource optimally with a network lag of 131.65 milliseconds and a network freeze of 0.09% with the resources utilized. This is evidence that offloading and rate control in combination provides better resource elasticity for M-IoT systems.