International Journal of Artificial Intelligence and Soft Computing (5 papers in press)

Regular Issues

• Enhancing Class Separability in Imbalanced Learning: a Novel Model and Comparative Study  
  by Eric Jiang 
  Abstract: Class imbalance and other data complexity issues such as class overlap are common in many real-world applications. These collective challenges can lead to biased learning models and reduced predictive performance. Various strategies that combine data sampling with data cleaning have been developed to address these challenges by mitigating learning bias and enhancing class separability. This paper introduces a model-based selective under-sampling approach designed to identify and remove potentially noisy or unreliable instances and rebalance the class distribution of training data. In addition, the paper conducts a comprehensive comparative study on several widely used selective under-sampling methods and it involves extensive experiments on a diverse collection of 40 datasets from a broad range of applications. The study performs in-depth comparisons of the examined methods using a framework of non-parametric statistical tests and provides valuable empirical findings, insights and observations that may benefit machine learning researchers and practitioners.
  Keywords: class distribution; class overlap; data sampling; data cleaning; non-parametric statistical test.
  DOI: 10.1504/IJAISC.2026.10078600
   
  
• Enhanced Real-Time Face Mask Detection Using Optimized SSD MobileNetV2 on NVIDIA Jetson Nano  
  by Priyanka Maan, Akanksha Sehrawat, Tanya Rana, Muskan Verma, S.R.N. Reddy, Rishika Anand 
  Abstract: The COVID-19 pandemic emphasized the importance of face masks in reducing airborne disease transmission. Automated face mask detection systems can support health regulations and improve safety in crowded areas. This paper proposes a real-time face mask detection framework based on an Enhanced SSD-MobileNetV2 model deployed on the NVIDIA Jetson Nano platform. The conventional model is improved using transfer learning, selective fine-tuning, data augmentation, confidence threshold optimization, and Soft Non-Maximum Suppression (Soft-NMS) to enhance accuracy and robustness. The model was trained using a Kaggle dataset containing mask, no-mask, and improper-mask classes, along with a custom real-time webcam dataset. Experimental results show an overall accuracy of 86.7% with low inference latency suitable for real-time deployment. The system effectively detects mask compliance under varying lighting, pose, and background conditions. Its lightweight architecture makes it suitable for smart surveillance, workplaces, transportation hubs, educational institutions, and healthcare facilities, offering a cost-effective public safety solution.
  Keywords: Jetson Nano; SSD MOBILENET V2; Object Detection; Face Mask Detection.
  DOI: 10.1504/IJAISC.2026.10079564
   
  
• A pessimistic dissimilarity measure on Fermatean fuzzy environment and its application in air quality ranking problems using MCDM algorithm  
  by Kaushik Kr. Sarmah, Brindaban Gohain, Chinmayee Devi, Rituparna Chutia 
  Abstract: Fermatean fuzzy sets have emerged as an effective mathematical framework for modelling high degrees of uncertainty in complex decision-making problems. Motivated by the need to incorporate risk-sensitive evaluation in such environments, this study proposes a novel Fermatean fuzzy dissimilarity measure based on a pessimistic viewpoint, in which non-belongingness information is assigned greater influence during similarity assessment. The mathematical properties of the proposed measure are established, and its performance is examined through comparative analysis with existing Fermatean fuzzy distance measures, demonstrating improved discrimination under uncertain conditions. The practical applicability of the proposed approach is illustrated through an air quality indexing and ranking problem of the ten most populous Indian cities using a Fermatean fuzzy multi-criteria decision-making framework. Both subjective and objective criteria weights are incorporated, followed by sensitivity analysis to assess ranking stability. The results indicate that the proposed dissimilarity-based multi-criteria decision-making model produces more cautious and robust ranking outcomes, making it particularly suitable for environmental and public health-oriented decision-making problems.
  Keywords: Dissimilarity measure; Fermatean fuzzy sets; pessimistic viewpoint; multi-criteria decision-making.
  DOI: 10.1504/IJAISC.2026.10079758
   
  
• A Hybrid Binary Salp Swarm Algorithm Approach for addressing the Maximum Clique Problem in Social Networks  
  by Amit Kumar 
  Abstract: This paper presents a Hybrid Binary Salp Swarm Algorithm (HBSSA) aimed at addressing the Maximum Clique Problem (MCP) in social networks. The approach begins with the formulation of a binary Salp Swarm Algorithm (BSSA) that incorporates a breakout local search (BLS) heuristic, creating a hybrid methodology. The performance of HBSSA is evaluated using publicly available DIMACS benchmark instances and the BHOSLIB dataset. The results demonstrate that HBSSA consistently identifies larger average cliques and surpasses other metaheuristic algorithms, including Bat Algorithm (BA), Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), and Harmony Search (HS), in terms of average execution time. This demonstrates the efficiency and effectiveness of the HBSSA in solving the MCP.
  Keywords: Salp swarm algorithm; Maximum Clique Problem; Breakout Local Search; Metaheuristics.
  DOI: 10.1504/IJAISC.2026.10080693
   
  
• OptimiSed Network Security Prediction Using SSA-Tuned GResNeSt and Variational Mode Decomposition  
  by Saravanan Krishnan, Antony Shyamalin Antony Pudumai Anandam, Srewathi TK, Ramya GK 
  Abstract: Network security situation prediction (NSSP) plays a crucial role in allowing preemptive detection of threats and efficient response to cyberattacks at appropriate times. Current models often experience difficulties in the effective handling of the dynamic, unstable, and nonlinear character of network security information. Herein, this paper present an innovative hybrid scheme that merges variational mode decomposition (VMD), CNN-bidirectional LSTM-attention mechanism, and global residual neural network with split attention (GResNeSt) to boost NSSP greatly. VMD is applied for decomposing the raw network traffic data into stationary intrinsic mode functions, reducing the noise and the unreliability of predictions. The CNN-BiLSTM-attention module preserves spatial and temporal relationships in the decomposed signals, whereas GResNeSt enhances feature learning by capturing cross-feature interactions and long-range relationships. To improve the models performance further, the salp swarm algorithm (SSA) (Mirjalili et al., 2017) is utilised for hyperparameter optimisation. The proposed approach is tested on benchmark cybersecurity datasets, such as HIKARI-2021 and Edge-IIoTset. Experimental results prove that our method surpasses traditional deep learning models in accuracy, efficiency, and robustness. Our work presents a scalable and smart solution to real-time network threat prediction for better situational awareness and proactive cyber defenCe.
  Keywords: Network security; Situation prediction; Variational mode decomposition; Deep learning; Attention mech mechanism; Hyperparameter optimization.
  DOI: 10.1504/IJAISC.2026.10080694