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

Regular Issues

• Large Language Models: State-of-the-Art Techniques, Applications and Emerging Trends  
  by M.Deepadharshana Moorthy, Vijayarani S 
  Abstract: An advanced artificial intelligence system called a large language model (LLM) is made to understand and provide language that resembles that of human beings. Built on deep learning models, particularly transformer architecture, LLMs leverage massive datasets and sophisticated neural network structures to excel in different types of language-based tasks. These models employ essential components such as self-attention mechanisms, embedding layers, and dense feed-forward layers to collect the syntactic and semantic fine details of language. Through extensive pre-training on diverse text corpora, LLMs learn intricate linguistic patterns, enabling them to execute jobs like translation, summarisation, accurate question answering, and text completion. LLMs are crucial in domains like content creation, customer service, and scientific research. The transformer architecture uses self-attention and feed-forward layers to process sequential data efficiently without recurrence. This architecture is widely used in natural language processing operations like text generation and translation.
  Keywords: Keywords: LLM; BERT; ANN; Data Annotation; Transformer; Self-Attention.
  DOI: 10.1504/IJAISC.2025.10072831
   
  
• Explainable AI for Deepfake Detection: Strategies to Mitigate Bias and Enhance Trustworthiness  
  by Neenu Maria Thankachan, Greeshma K. V, Archana Sunil, Binshad M. S 
  Abstract: Deepfakes, hyper-realistic synthetic media, pose significant challenges by depicting individuals in situations they never experienced. While AI-powered deepfake detection methods show promise, they are prone to bias, affecting the authenticity and fairness of outcomes. This study explores the role of Explainable AI (XAI) in reducing bias in deepfake detection systems. We analyse bias sources, including data and algorithmic biases, and propose strategies to mitigate them using XAI techniques like LIME and SHAP. We demonstrate how XAI enhances transparency, facilitates bias detection, and promotes fairer deepfake detection outcomes. Our research underscores the ethical implications of AI and the need for trustworthy AI systems to ensure societal well-being. This work contributes to the development of more effective and trustworthy deepfake detection mechanisms.
  Keywords: Artificial Intelligence; Datasets; Deepfake; XAI; AI-generated images; Generative AI.
  DOI: 10.1504/IJAISC.2025.10073240
   
  
• Comparative Analysis of Arabic Transformers Using an Arabic Fake News Dataset  
  by Amine Mammasse, Khaled Bedjou, Faical Azouaou 
  Abstract: This study presents a comparative analysis of twelve state-of-the-art Arabic Transformer models, addressing the challenges of fake news detection in Arabic Natural Language Processing (NLP). We evaluate models including AraT5v2-base-1024, MARBERT, AceGPT, and ArabianGPT using a unified corpus integrating three publicly available datasets (AFND, ANS, and AraFacts). Our novelty lies in systematically analysing how model architecture, pre-training data quality, and fine-tuning strategies influence performance across Arabic dialectal variations and linguistic complexities. Results demonstrate ArabianGPT achieving superior performance with an F1-score of 82.4%, followed by PhoenixMultiple-Langs-v1 and BERT-base-QARIB. The findings reveal decisive correlations between detection accuracy and factors such as architectural design, training data diversity, and dialectal adaptability. This pioneering analysis uncovers critical insights into the importance of morphological richness, script-specific preprocessing, and comprehensive dialect coverage in Arabic NLP. Our research underscores the need for more diverse training data and specialized architectures to advance technologies combating misinformation across Arabic-speaking regions.
  Keywords: Arabic Natural Language Processing; Transformer Models; Fake News Detection; Arabic Datasets; Neural Networks.
  
  
• Enhanced Android Security with Real-time Cloud-Driven Malware Detection and Multi-Model Ensemble Predictions  
  by Isaac Osunmakinde 
  Abstract: This study addresses the growing sophistication of cyber threats targeting Android devices. This demands more robust security systems. Existing solutions often struggle with high false alarm rates and limited adaptability to novel threats. This research proposes a multi-model ensemble framework that integrates Decision Trees (DT), Random Forests (RF), and Neural Networks (NN) for real-time predictive threat analytics. The system incorporates data pre-processing techniques (log transformation and SMOTE) to address data imbalances and utilises iterative feature selection to optimise performance. Experimental validation on Android traffic and zero-day ransomware datasets demonstrates the model's effectiveness. The ensemble model achieves 98.59% accuracy on Android malware detection with a 1.49% false alarm rate, outperforming individual models and previous studies. The framework achieves 99.41% test accuracy for zero-day threat detection, with a false alarm rate of only 0.51%, confirming its high generalisability. A cloud-based setup enables ongoing, scalable, and proactive cybersecurity for Android and other platforms.
  Keywords: Android Security; Ensemble Model; Malware Detection; Real-time; Cloud-Driven; Cybersecurity; Machine Learning.
  
  
• Studying the Stock Market Behaviour under the Lens of Crude Oil Price and Exchange Rates  
  by Raktim Ghosh, Ashis Kumar Sana, Bhaskar Bagchi 
  Abstract: This study focuses on the stock market behaviour of the fragile five nations using the Multilayer Perceptron model under the Artificial Neural Network through crude oil price and exchange rates. Though the concept of the Fragile Five was developed in the year 2013, keeping in mind the occurrence of the Global financial recession of 2008, which is also the foundation of the Fragile Five, it is decided to consider the study period from January 1, 2008, to August 1, 2024. It is worth reporting that the exchange rate has greater relevance in terms of predictability. Uniqueness lies in studying amid the COVID-19 pandemic, Russia-Ukraine crisis, and other global happenings offering innovative insights. In the context of global turmoil, the fragile five nations face increased vulnerability, impacting energy costs, commodity prices, currency devaluation, and the global supply chain. MLP model features are tailored to address the challenges posed by these economies.
  Keywords: Stock Market; Fragile Five; Artificial Neural Network; Crude oil prices; Exchange Rates.