International Journal of Computer Science and Technology

International Journal of Computer Science and Technology | Vol. 2, No. 12, December 2021 | pp. 89–96

Research Article

Title: A Multi-Criteria Task Scheduling Framework for Heterogeneous Cloud Systems Using Simulated Annealing and Ant Colony Optimization

Names of Authors: Kofi Mensah¹, Kwame Boateng², and Ama Asare³

Authors’ Affiliations:
¹Department of Computer Science, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
²Department of Computer Engineering, University of Ghana, Accra, Ghana
³School of Technology, GIMPA, Accra, Ghana

Abstract: Efficient task scheduling remains a critical operational priority for infrastructure-as-a-service cloud datacenters seeking to optimize system performance while reducing resource overheads. The assignment of diverse, inter-dependent computational tasks onto non-uniform physical servers represents a complex, multi-objective NP-hard problem. Standard scheduling heuristics usually prioritize single metrics like overall completion time, which frequently leads to uneven server load distribution, high energy waste, and service level agreement violations. This study proposes an optimized hybrid metaheuristic framework, SA-ACO, combining Ant Colony Optimization (ACO) with Simulated Annealing (SA) to resolve complex cloud scheduling demands. The algorithm utilizes an initial ACO loop to establish global routing routes across available nodes, while an integrated SA mutation routine alters local node assignments to avoid getting trapped in local optima. The objective model evaluates multiple performance trade-offs, factoring in task completion time, resource utilization balance, and server energy consumption. Extensive benchmark simulations with 1000 tasks running across 100 virtual machines reveal that the SA-ACO framework lowers overall completion time by 21.4% and decreases system energy consumption by 18.7% compared to traditional genetic algorithms and standard round-robin scheduling setups.

Keywords: Cloud computing, Task scheduling, Ant colony optimization, Simulated annealing, Resource allocation, Energy optimization

Manuscript Timeline: Received: September 15, 2021; Revised: October 24, 2021; Accepted: November 19, 2021; Published: December 1, 2021

Citation: Mensah, K., Boateng, K., & Asare, A. (2021). A multi-criteria task scheduling framework for heterogeneous cloud systems using simulated annealing and ant colony optimization. International Journal of Computer Science and Technology, 2(12), 89–96. DOI: 10.46882/2021/IJCST/000024

International Journal of Computer Science and Technology | Vol. 2, No. 11, November 2021 | pp. 81–88

Research Article

Title: An Automated Malicious Domain Detection Architecture Using Recurrent Neural Networks and Passive DNS Analysis

Names of Authors: Youssef Mansour¹, Hassan Farhat², and Fatima Al-Sayed³

Authors’ Affiliations:
¹Department of Computer Science, American University of Beirut, Beirut, Lebanon
²Faculty of Technology, Lebanese University, Sidon, Lebanon
³Department of Computer Engineering, Beirut Arab University, Beirut, Lebanon

Abstract: Cybercriminals regularly utilize algorithmic techniques like Domain Generation Algorithms (DGAs) to generate vast lists of temporary internet domains for command-and-control communication. Because these domains change rapidly, conventional reputation blacklists struggle to provide real-time protection, leaving enterprise networks vulnerable to ransomware and data extraction campaigns. This paper introduces an automated, high-precision malicious domain detection system that combines structural character analysis with passive DNS traffic modeling. The core architecture implements a long short-term memory (LSTM) recurrent neural network that checks string sequences to flag algorithmically generated domains before they receive active network requests. Simultaneously, a statistical profiling layer monitors passive DNS telemetry, assessing attributes like query frequency variations, geographic IP distributions, and resource record changes. Training and validation leverage a large dataset of 500,000 active domains, spanning both legitimate entries and real-world botnet traffic. Experimental metrics demonstrate an overall detection accuracy of 98.9% alongside a low false positive rate of 0.014%. The integrated detection engine processes incoming queries in less than 3.5 ms, making it highly effective for deployment within active corporate DNS gateways to block zero-day command-and-control communication channels.

Keywords: Network security, Domain generation algorithms, Long short-term memory, Passive DNS, Intrusion prevention, Cyber forensics

Manuscript Timeline: Received: August 18, 2021; Revised: September 22, 2021; Accepted: October 14, 2021; Published: November 1, 2021

Citation: Mansour, Y., Farhat, H., & Al-Sayed, F. (2021). An automated malicious domain detection architecture using recurrent neural networks and passive DNS analysis. International Journal of Computer Science and Technology, 2(11), 81–88. DOI: 10.46882/2021/IJCST/000023

International Journal of Computer Science and Technology | Vol. 2, No. 10, October 2021 | pp. 73–80

Research Article

Title: A Scalable Sharding Protocol with Dynamic Cross-Shard Transaction Verification for High-Throughput Blockchains

Names of Authors: Tan Nguyen¹, Hanh Tran², and Minh Le³

Authors’ Affiliations:
¹Faculty of Computer Science, University of Information Technology, Ho Chi Minh City, Vietnam
²Department of Information Technology, Vietnam National University, Hanoi, Vietnam
³School of Computer Engineering, Hanoi University of Science and Technology, Hanoi, Vietnam

Abstract: Blockchain systems encounter severe scalability barriers due to the constraint that every node must validate every network transaction. Sharding protocols resolve this bottleneck by splitting the network ledger into smaller, parallel processing units called shards. However, current sharding architectures generate heavy communication overhead and elevated transaction latency when executing cross-shard transactions, which compromises overall network throughput. This paper introduces an optimized, secure sharding protocol that utilizes a dynamic cross-shard validation pipeline driven by a two-phase Byzantine Fault Tolerance (BFT) consensus mechanism. The proposed framework implements a lock-free state atomic swap engine, allowing separate shards to verify mutual transaction dependencies simultaneously without stalling local ledger validation. To defend against 51% takeover exploits within individual sub-shards, a verifiable random function (VRF) executes continuous, unpredictable node re-allocation across the network topology. Performance evaluations conducted over a simulated setup of 2000 nodes distributed across 16 shards show a sustained transaction processing capacity of 8500 transactions per second (tx/sec). This performance represents a 42.6% throughput improvement over existing static sharding solutions. Furthermore, cross-shard confirmation latency drops by 31.2%, proving the protocol's viability for high-capacity enterprise ledger networks.

Keywords: Blockchain scalability, Network sharding, Consensus protocols, Byzantine fault tolerance, Cross-shard transactions, Verifiable random function

Manuscript Timeline: Received: July 11, 2021; Revised: August 19, 2021; Accepted: September 15, 2021; Published: October 1, 2021

Citation: Nguyen, T., Tran, H., & Le, M. (2021). A scalable sharding protocol with dynamic cross-shard transaction verification for high-throughput blockchains. International Journal of Computer Science and Technology, 2(10), 73–80. DOI: 10.46882/2021/IJCST/000022

International Journal of Computer Science and Technology | Vol. 2, No. 9, September 2021 | pp. 65–72

Research Article

Title: An Optimized Deep Learning Framework for Real-Time Facial Expression Recognition in Neurological Assessment

Names of Authors: Elena Rostova¹, Dmitry Ivanov², and Anna Kuznetsova³

Authors’ Affiliations:
¹Department of Computer Science, National Research University Higher School of Economics, Moscow, Russia
²Faculty of Computational Mathematics and Cybernetics, Lomonosov Moscow State University, Moscow, Russia
³Institute of Artificial Intelligence, Saint Petersburg State University, Saint Petersburg, Russia

Abstract: Automated facial expression recognition (FER) systems have emerged as powerful non-invasive tools for monitoring patient neurological responses and cognitive disorders in clinical settings. However, applying FER models to clinical diagnostic workflows requires ultra-low latency processing and robust adaptation to irregular lighting and head pose variations. Standard deep neural networks often suffer from high computational overhead and drop in classification accuracy when deployed on lightweight clinical monitoring units. This paper presents an optimized, real-time FER framework utilizing a modified MobileNetV3 architecture combined with a coordinate attention mechanism. The proposed system retains spatial context by embedding positional info into channel attention, enhancing fine-grained micro-expression extraction. To combat dataset bias, the architecture undergoes transfer learning using a combined dataset of AffectNet and clinical facial sequences. Quantitative evaluations show that the proposed system achieves a classification accuracy of 89.6% across seven core emotional states while restricting model parameter size to 4.2 megabytes (MB). Hardware profiling on an embedded edge device yields an average inference speed of 38.5 frames per second (fps). The framework delivers an efficient, reliable solution for continuous patient monitoring during neurological evaluations without requiring high-end computational infrastructure.

Keywords: Facial expression recognition, Edge intelligence, Deep learning, MobileNet, Coordinate attention, Neurological monitoring

Manuscript Timeline: Received: June 15, 2021; Revised: July 20, 2021; Accepted: August 14, 2021; Published: September 1, 2021

Citation: Rostova, E., Ivanov, D., & Kuznetsova, A. (2021). An optimized deep learning framework for real-time facial expression recognition in neurological assessment. International Journal of Computer Science and Technology, 2(9), 65–72. DOI: 10.46882/2021/IJCST/000021

International Journal of Computer Science and Technology | Vol. 2, No. 8, August 2021 | pp. 57–64

Research Article

Title: An Automated Smart Contract Vulnerability Detection Framework Using Static Analysis and Graph Neural Networks

Names of Authors: Yuki Nakamura¹, Aiko Tanaka², and Daiki Sato³

Authors’ Affiliations:
¹Department of Computer Science, Tokyo Institute of Technology, Tokyo, Japan
²Graduate School of Information Science and Technology, Osaka University, Osaka, Japan
³Department of Information Engineering, Nagoya University, Nagoya, Japan

Abstract: Smart contracts deployed on public blockchain platforms manage billions of dollars in digital assets, making them high-stakes targets for cyber exploits. Structural vulnerabilities such as reentrancy, integer overflows, and timestamp dependencies have historically led to catastrophic capital losses. Traditional vulnerability identification relies on manual auditing or rule-based static symbolic execution, both of which struggle with high false-positive rates and fail to uncover deep semantic bugs in complex contract control flows. This paper proposes a deep-learning-driven smart contract vulnerability detection framework that combines static analysis with Graph Neural Networks (GNNs). The system compiles raw Solidity smart contract source code into abstract syntax trees (ASTs) and control flow graphs (CFGs), which are then merged into a unified Code Property Graph (CPG) representing data dependencies and execution paths. A Bidirectional Gated Graph Neural Network (BGGNN) processes these graphs to learn vector embeddings of the contract's structural features. Experimental evaluations on a curated dataset of 45,000 smart contracts demonstrate that the proposed framework identifies critical vulnerabilities with an overall accuracy of 96.4% and an F1-score of 0.958, outperforming conventional tools like Oyente, Mythril, and standard recurrent neural networks. The system provides an efficient, automated security validation utility for smart contract deployment pipelines.

Keywords: Smart contract security, Blockchain, Graph neural networks, Static analysis, Vulnerability detection, Deep learning

Manuscript Timeline: Received: May 12, 2021; Revised: June 19, 2021; Accepted: July 15, 2021; Published: August 1, 2021

Citation: Nakamura, Y., Tanaka, A., & Sato, D. (2021). An automated smart contract vulnerability detection framework using static analysis and graph neural networks. International Journal of Computer Science and Technology, 2(8), 57–64. DOI: 10.46882/2021/IJCST/000020

International Journal of Computer Science and Technology | Vol. 2, No. 7, July 2021 | pp. 49–56

Research Article

Title: A Software-Defined UAV Swarm Communication Network Utilizing Multi-Agent Reinforcement Learning for Dynamic Routing

Names of Authors: Sanjay Nair¹, Priya Pillai², and Anand Krishnan³

Authors’ Affiliations:
¹Department of Aerospace Engineering, Indian Institute of Science, Bangalore, India
²Department of Computer Science and Automation, Indian Institute of Science, Bangalore, India
³Department of Electronics and Communications, National Institute of Technology, Calicut, India

Abstract: Unmanned Aerial Vehicle (UAV) swarms are increasingly deployed in disaster monitoring, environmental surveillance, and search-and-rescue operations due to their mobility and flexible deployment capabilities. However, maintaining reliable ad-hoc communication links within a flying swarm is difficult because high-speed three-dimensional node mobility causes frequent topological disruptions and link disconnects. Traditional routing protocols face scalability issues and slow convergence, leading to severe packet drops and routing loop overheads. This research introduces a software-defined UAV swarm networking architecture that utilizes a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm for adaptive routing control. The framework uses a centralized Software-Defined Networking (SDN) controller on a ground base station to maintain global topology states, while individual UAV nodes run localized packet forwarding policies. The multi-agent learning algorithm models routing as an optimization problem where each UAV learns to select the optimal next-hop neighbor based on node positions, residual energy levels, link qualities, and packet queue constraints. Simulation experiments conducted in dynamic 3D flying corridors show that the MADDPG routing protocol reduces end-to-end delay by 39.4% and improves the overall packet delivery ratio by 22.8% over traditional Ad-hoc On-Demand Distance Vector (AODV) and Optimized Link State Routing (OLSR) protocols. The system ensures robust link stability in high-mobility deployment contexts.

Keywords: UAV swarms, Software-defined networking, Multi-agent reinforcement learning, Dynamic routing, Ad-hoc networks, Wireless communication

Manuscript Timeline: Received: April 15, 2021; Revised: May 22, 2021; Accepted: June 16, 2021; Published: July 1, 2021

Citation: Nair, S., Pillai, P., & Krishnan, A. (2021). A software-defined UAV swarm communication network utilizing multi-agent reinforcement learning for dynamic routing. International Journal of Computer Science and Technology, 2(7), 49–56. DOI: 10.46882/2021/IJCST/000019