The Indonesian Journal of Computer Science

Indicators and Measurement Methods for Smart City Performance: A Literature Review

Ghefira Nur Fatima, Muhayat, Siti Alayda Azzahro — Tue, 09 Jun 2026 00:00:00 +0700

This study examines the evolution of smart city performance measurement, focusing on the development of concepts, the types of indicators applied, the measurement methods used, and the challenges found in this field through a comprehensive review of previous studies. The research analyzes 25 journal articles indexed in Scopus and SINTA that discuss topics such as smart city indicators, composite indices, Internet of Things (IoT), network Quality of Service (QoS), resilience, cybersecurity, and governance. The data were analyzed using thematic-comparative analysis and configurational analysis to identify conceptual patterns and methodological differences across the studies. The findings indicate a shift from technology-centered measurement models toward more comprehensive frameworks that incorporate multiple dimensions, including environmental, economic, mobility, governance, social, and digital infrastructure aspects. Differences in data normalization, weighting, index aggregation, and IoT-based data collection influence the accuracy and interpretability of performance assessments. The study also highlights the need to integrate cybersecurity, system reliability, and stakeholder collaboration to develop more robust and context-appropriate smart city evaluation frameworks.

Hybridized Machine Learning based IDS for Anomaly Detection: A Systematic Review

Victor Mathebula, Bukohwo Michael Esiefarienrhe — Mon, 25 May 2026 00:00:00 +0700

Intrusion Detection Systems play a crucial role in safeguarding networks against increasingly sophisticated cyber threats. Traditional Intrusion Detection Systems approaches often struggle with adaptability and high false-positive rates. This review investigates the use of hybridized Machine Learning models for anomaly detection in IDS to enhance detection accuracy and system robustness. This study applies the PRISMA framework to analyze hybrid machine learning techniques applied to improve the performance of Intrusion Detection Systems, the datasets used, performance evaluation, identification of challenges, and knowledge gap analysis. Results show that hybrid ML models consistently outperform single-model approaches, achieving an accuracy of up to 99.99%. Despite promising results, challenges such as class imbalance and limited real-time deployment persist. From this systematic review, it is evident that hybridizing machine learning algorithms in Intrusion Detection Systems offers a powerful approach to anomaly detection, improving precision and accuracy.

A Novel Deep Learning Framework for Biomedical and Medical Image Classification with Adaptive Feature Learning Approach and Enhanced Diagnostic Performance

Mohanaed Ajmi Falih — Mon, 25 May 2026 00:00:00 +0700

Owing to the significance of early diagnosis and proper clinical decision-making, medical image analysis has now become an essential part of modern healthcare systems. Traditional image analysis techniques weaken for complex patterns, variability in medical data and a high diagnostic accuracy need We therefore propose in this paper a new deep learning-based framework for adaptive feature learning and improved diagnostic accuracy. We then present a convolutional neural network (CNN) based hybrid architecture framed with an adaptive feature learning strategy that learns suitable spatial and semantic characteristics for fine-tuning dynamically. The proposed method will help the model more accurately discover small details in medical images (e.g., lesions, tumors and unusual tissue structures). In addition, a feature fusion strategy is applied to combine multiscale representations which improve the robustness of multiple imaging modalities (MRI, CT and X-ray). Besides that the model implements various optimizations: batch norm, dropout regularization and adaptive learning rate scheduling. Extensive experiments conducted on benchmark medical imaging datasets confirmed the efficacy of the proposed method. The result shows that it outperforms any previous existing state-of-the-art methods by a considerable margin on the accuracy, precision, recall and f1-score metrics. This framework also obtained a greater performance on generalizability, and decreased sensitivity to noise-induced fluctuations over quality of picture changes. We demonstrate that the incorporation of tray-like adaptive feature learning through deep neural networks can lead to substantial improvements on image such as medical imaging. This paper provides a foundation for the development of intelligent, trustworthy and scalable AI based health care systems that assist clinicians in improving decision making with high correct classification rates as fast as possible.

Performance Analysis of Vision Transformer (ViT), ResNet50, and MobileNetV3 Large in Multiclass Bone Fracture Classification

Ei Phyu Sin Win, Phyo Thu Zar Tun — Mon, 01 Jun 2026 00:00:00 +0700

Automated classification of bone fractures has become a cornerstone of modern emergency radiology, significantly enhancing diagnostic speed and precision. This study evaluates the comparative efficacy of three leading deep learning frameworks ResNet50, MobileNetV3, and Vision Transformer (ViT) using a diverse dataset that includes various fracture modalities, healthy X-rays, and non-radiological images.The experimental data reveals that the Vision Transformer (ViT) attained the highest diagnostic accuracy at 95%, marginally outperforming MobileNetV3 and ResNet50, which both achieved 94%. While all three models demonstrated flawless reliability (100%) in identifying Forteen Classes Bone categories, their performance diverged when analyzing complex fracture patterns.

A Multi-View Anomaly Detection Framework for Elephant Movement Based on GPS Data

Mon, 01 Jun 2026 00:00:00 +0700

Monitoring elephant movement is crucial for wildlife conservation, especially under threats such as poaching and habitat loss. With the availability of large-scale GPS tracking data, anomaly detection can help identify abnormal behaviors linked to critical events. However, challenges such as data imbalance, GPS noise, and real-time deployment constraints remain. This paper proposes an end-to-end framework for anomaly detection in elephant movement using GPS data. The approach combines multi-view anomaly modeling with a weighted scoring mechanism and a lightweight Random Forest model. To address class imbalance, the pipeline integrates SMOTE (Synthetic Minority Over-sampling Technique), under sampling, and class-weighted learning. Feature selection and quantization further optimize the system for edge and FPGA deployment. Experimental results show strong performance, with F1-Macro ≈ 0.98, ROC-AUC ≈ 0.99, and high recall for anomaly detection. The proposed framework provides an efficient and practical solution for real-time wildlife monitoring.

An Extensive Analysis and Taxonomy of Explainable Artificial Intelligence for Malware Identification

Dauan Aziz, Firas Mohammed Amien, Raghad Zuhair Yousif — Tue, 09 Jun 2026 00:00:00 +0700

As malware continues to evolve in sophistication and scale, traditional detection methods struggle to keep pace, especially when facing obfuscated or zero-day threats. In response, Machine Learning (ML) and Deep Learning (DL) techniques have shown significant promise in enhancing malware detection through pattern recognition and anomaly classification. However, their increasing complexity has introduced major interpretability challenges, particularly in high-stakes cybersecurity contexts. This paper provides a comprehensive survey of eXplainable Artificial Intelligence (XAI) methods applied to malware detection across diverse computing platforms, including Windows PE files, PDF, Linux, and hardware-based systems. We propose a novel taxonomy that categorizes explainable malware detection approaches by model transparency, explanation technique (model-agnostic or model-specific), and deployment environment. We also discuss major trends, highlight underexplored domains, and outline future research directions aimed at enhancing real-time interpretability, adversarial robustness, and human-in-the-loop integration. This work aims to bridge the gap between high-performance malware detection models and actionable, transparent security decision-making.

Design and Performance Test of Squirrel Cage Hydrokinetic Turbine

Yin Yin Aye — Tue, 09 Jun 2026 00:00:00 +0700

This paper focuses on the design and performance test of squirrel cage hydrokinetic turbine. The design water flow velocity is 1.2 m/s and the designed water turbine has 3 blades. Turbine design is calculated by changing the aspect ratio. According to design calculation, maximum torque is found on aspect ratio 0.5. Squirrel cage hydrokinetic turbine is constructed by using the calculated design data. The turbine design diameter is 0.925m and height is 0.463 m. The turbine performance is tested at a canal near Mandalay Technological University. The experimental test results of rotational speed, angular velocity and turbine power for water velocity 1.2 m/s are 29 rpm, 3 rad/s and 68.2 W respectively. Moreover, theoretical and experimental test results of rotational speed, angular velocity and turbine power are compared by changing water velocity. According to the theoretical and experimental results, if the water velocity is increased, the rotational speed and turbine power are gradually increased.

Keywords :

aspect ratio, performance test, power, theoretical, water velocity

Treatment of Textile Wastewater with Activated Carbon Produced from Plum Seed Shell

Mon Yi Myo Myint, Zin Marlar Tin San, Nway Nway Khaing — Tue, 09 Jun 2026 00:00:00 +0700

In this study, the activated carbon derives from plum seed shells are used for the treatment of textile wastewater collected from the Wandwin area which has high concentrations of parameters such as pH, colour, TSS, TDS, COD, and BOD. pH value of textile wastewater is 12.4 that leads to alkaline, so pH adjusts with alum coagulants. After treatment the COD concentration can be reduced from 1280 mg/l to 20 mg/l by using equilibrium concentration of activated carbon, C_e=50 mg/. The adsorption data followed the Langmuir isotherm (R² = 0. 0.97832). Other parameters also significantly decreased from 74.74 % to 98.29%. According to the National Enviromental Quality Guideline (NEQG) Myanmar, all treated water parameters comply with NEQG standards. Therefore, the plum seed shell activated carbon is an effective and sustainable material for the treatment of textile wastewater and treated effluent can be safely discharged into the surrounding water bodies.

A Hybrid Deep Learning Framework for Malware Detection Using Metaheuristic Feature Selection and Explainable AI: A Comprehensive Literature Review

Dauan Aziz, Firas Mohammed Amien, Raghad Zuhair Yousif — Tue, 09 Jun 2026 00:00:00 +0700

Malware, including ransomware, trojans, rootkits, spyware, and advanced persistent threats (APTs), poses a growing challenge to modern computing systems. Traditional detection methods, such as signature- and heuristic-based approaches, struggle to detect polymorphic, metamorphic, and zero-day malware. Deep learning has emerged as a powerful solution due to its ability to automatically learn hierarchical features. However, key challenges remain: lack of model transparency, high-dimensional feature redundancy from multimodal analysis, and poor cross-dataset generalization. This paper presents a systematic literature review of state-of-the-art malware detection techniques published between 2020 and 2025, covering static, dynamic, visualization-based, and deep learning approaches (e.g., CNN, LSTM, BiLSTM, and hybrid models), along with metaheuristic feature selection and Explainable AI (XAI) methods such as SHAP, Grad-CAM, and LIME. Analysis of 35 studies identifies critical gaps, particularly the absence of integrated metaheuristic optimization and XAI-driven hybrid frameworks, motivating future research directions.

Optimization of Charging and Discharging Performance in a PV-Integrated Piston-Based Gravity Energy Storage System

Khin Mar Myint, Wunna Swe — Tue, 16 Jun 2026 00:00:00 +0700

The rapid integration of photovoltaic systems into modern power networks has created operational challenges due to their intermittent and fluctuating output. To maintain power balance and enhance grid stability, large-scale and efficient energy storage solutions are essential. This paper presents the dynamic modeling and simulation of a piston-based gravity energy storage system integrated with PV power plant. A detailed MATLAB/Simulink model is developed, including the PV array, power electronic converters, motor–pump system, piston-based gravitational storage mechanism, and generator–load interface. During periods of surplus PV generation, the system operates in charging mode by driving the motor–pump unit to lift the piston, thereby storing energy as gravitational potential. During low PV output, the stored energy is released through the turbine–generator unit to support the load. Simulation results verify stable electrical performance, smooth transition between operating modes, and effective mitigation of PV power fluctuations, demonstrating the technical feasibility of the proposed PGES configuration.

Impact of 5G in Agricultural Networks: A Review of Improvement Strategies and Publication Trends

Alfred Kgopa — Mon, 08 Jun 2026 00:00:00 +0700

This study presents a systematic literature review of the impact of fifth-generation (5G) wireless networks on agriculture, focusing on publication trends, challenges, and opportunities from Scopus papers published between 2015 and 2025. The findings reveal a rapid annual growth rate of 34.93% in research outputs, with 274 documents published across 96 sources, indicating rising global interest in 5G-enabled smart farming technologies. However, persistent challenges include infrastructural limitations in rural farm areas, high costs of implementation, and digital literacy gaps among smallholder farmers. The bibliometric analysis highlights strong international collaboration (28.47%) and dominant contributions from India, China, and the USA. Emerging research trends focus on artificial intelligence (AI), edge computing, and hybrid connectivity models, signaling future directions for 5G-based agricultural innovations. Beyond mapping publication trends, this study synthesizes technical integration pathways and policy-relevant strategies for deploying resilient 5G-enabled agricultural networks in underdeveloped regions.

Ultra-Low Power Soil Sensor Enabling Multi-Year Battery Life

Haryono — Mon, 15 Jun 2026 00:00:00 +0700

Soil monitoring in large and remote agricultural areas is often limited by manual observation, resulting in low data frequency and reduced decision accuracy. To overcome this, this study presents a next-generation soil sensor system designed for ultra-low power operation and extended battery life. The system utilizes a STM32WLE5CCU6 microcontroller combined with a 6-in-1 RS485 soil sensor (T-H-EC-NPK) to measure soil humidity, temperature, electrical conductivity, and nutrient levels (N, P, K). The system operates using an optimized duty cycle, remaining in deep-sleep mode at approximately 3 μA and periodically activating for a short 600 ms sensing and data transmission phase consuming around 150 mA. This approach significantly reduces average power consumption to approximately 0.028 mA. With a 4000 mAh battery and a transmission interval of one hour, the system achieves a theoretical lifetime exceeding 16 years. However, considering practical factors such as battery self-discharge, sensor overhead, and environmental conditions, the effective operational lifetime is conservatively estimated to exceed 5–10 years without battery replacement. The results demonstrate that the proposed design successfully enables long-term, maintenance-free soil monitoring, making it suitable for large-scale and remote precision agriculture applications where energy efficiency and system reliability are critical.

Energy Management of Grid Connected PV-Wind-Battery Hybrid Power Supply System Using Model Predictive Control

Hnin Nwe, Wunna Swe — Mon, 15 Jun 2026 00:00:00 +0700

The increasing integration of renewable energy sources into modern power systems presents significant challenges in energy management due to their intermittent and uncertain nature. This paper proposes an energy management strategy for a grid connected photovoltaic (PV), wind and battery hybrid system based on Model Predictive Control (MPC). The main objectives are to enhance system efficiency, minimize grid power exchange and extend battery lifetime while ensuring reliable load supply. The proposed MPC-based EMS employs short-term forecasts of renewable generation and load demand to optimize power dispatch under system constraints. Its performance is evaluated and compared with a conventional rule-based EMS using MATLAB/Simulink. The results indicate that the MPC-based approach achieves smoother battery operation, improved state-of-charge regulation, enhanced renewable energy utilization and reduced grid dependency. Therefore, the proposed strategy provides on effective solution for advanced hybrid renewable energy systems.

The Minimal isiNdebele Explicit and Implicit Question-answering Text Datasets for Zero-shot Learning

Promise Malatji, Thipe Modipa — Tue, 30 Jun 2026 00:00:00 +0700

The use of Large Language Models (LLMs) for task specific has gained more popularity for resourced languages like English due to availability of text data. There are no readily available datasets to fine tune for task specific in low resourced languages like isiNdebele. Data scarcity remains the main obstacle for many low resourced languages as it hinders on the development of language models. In this study we proposed the creation of two isiNdebele Question-answering (QA) text datasets, for explicit and implicit QA. Forty-five matric question papers from the South African Department of Basic Education website were downloaded. A data verification process was performed using human-in-the-loop technique to validate the created datasets. Further data augmentation was performed to increase the implicit dataset from 3012 to 36560 context-question-answer triplets. The augmented implicit dataset was then used to fine-tune the mT5 model on a zero-shot. The two datasets were accepted with a 98.33% acceptance percentage by the participants. The mT5 model performed exceptionally with ROUGE-L and BERTScore F1 of 0,992 and 0,999 respectively. The model made 94.7% accurate predictions with a perplexity score of 1,157. The results indicate that the multilingual model and transfer learning have great potential of dealing with low resourced languages in such as QA.

A Comparative Analysis of Bi-LSTM and XGBoost for Time-Series Classification in Power System Stability using SMOTE and Focal Loss

Amal El Arid, Mahmoud Samad, Ghalia Nassreddine — Mon, 25 May 2026 00:00:00 +0700

The rapid deployment of renewable energy sources has raised several issues regarding grid stability. Renewable energy sources differ from traditional energy generation due to their volatile nature and dependence on climatic factors. Therefore, accurate predictions of smart grids' stability are essential for efficient energy management. Current approaches do not incorporate dynamic aspects into consideration. The purpose of this paper is to overcome this limitation through the development of a time series-based framework for smart grid stability detection. A bidirectional long short-term memory model is created to capture both short-term and long-term relationships of the grid frequency and power signal. A hybrid method is proposed combining the Synthetic Minority Oversampling Technique and focal loss to improve results for imbalanced datasets. An extreme gradient boosting model is trained based on flattened temporal sequences and statistical feature descriptions. The experimental findings indicate that the suggested framework demonstrates high predictive performance, with XGBoost achieving the best accuracy, while BiLSTM is effective for capturing temporal patterns and maintaining high stability in classification recall.