International Journal of Intelligent Systems and Applications (IJISA)

IJISA Vol. 17, No. 2, Apr. 2025

Cover page and Table of Contents: PDF (size: 200KB)

Table Of Contents

REGULAR PAPERS

An Optimized YOLOv8-based Method for Airport Bird Detection: Incorporating ECA Attention, MBC3 Module, and SF-PAN for Improved Accuracy and Speed

By Zia Ur Rehman Abdul Ghafar Ahmad Syed Abu Tayab Md. Golam Rabbi

DOI: https://doi.org/10.5815/ijisa.2025.02.01, Pub. Date: 8 Apr. 2025

This research investigation utilizes deep learning object detection algorithms to achieve accurate recognition of birds near airports, thereby addressing the limitations of manual bird detection at airports, including low accuracy slow speed, and the high cost of radar detection. The ultimate goal is to ensure the safe operation of civil aviation. The following are the primary enhancements: First, an ECA (Efficient Channel Attention) attention mechanism was added to the Neck to enhance the network's emphasis on important characteristics. This resulted in a notable improvement in accuracy while only changing a few parameters. Second, by adding branches with various receptive fields, the MBC3 (Muti Branch C3) module was created to improve the expressiveness of the model. Thirdly, the model's right width and depth parameters will be chosen by investigating the effects of various network widths and depths on model performance. Fourth, to solve the problem of feature loss in recognizing tiny bird targets, the SF-PAN (Shallow Feature - Path Aggregation Network) structure was proposed. The model was evaluated using metrics such as mAP@50, FPS, precision, recall, and computational complexity on a test set derived from the dataset. Results show that the enhanced YOLOv8 achieves a mAP@50 of 83.1% and a speed of 31 FPS, a 2.5% improvement in accuracy and a 7 FPS increase over the baseline YOLOv8, while reducing parameters and weight size by approximately 48%. Comparative experiments further validate the model’s superiority over existing algorithms in terms of accuracy and resource efficiency. This upgraded YOLOv8 provides a novel, real-time solution for precise bird detection in challenging airport environments, ensuring safer civil aviation operations.

[...] Read more.
Unveiling Hidden Patterns: A Deep Learning Framework Utilizing PCA for Fraudulent Scheme Detection in Supply Chain Analytics

By Kowshik Sankar Roy Pritom Biswas Udas Bashirul Alam Koushik Paul

DOI: https://doi.org/10.5815/ijisa.2025.02.02, Pub. Date: 8 Apr. 2025

Supply chain fraud, a persistent issue over the decades, has seen a significant rise in both prevalence and sophistication in recent years. In the current landscape of supply chain management, the increasing complexity of fraudulent activities demands the use of advanced analytical tools. Despite numerous studies in this domain, many have fallen short in exploring the full extent of recent developments. Thus, this paper introduces an innovative deep learning-based classification model specifically designed for fraud detection in supply chain analytics. To enhance the model's performance, hyperparameters are fine-tuned using Bayesian optimization techniques. To manage the challenges posed by high-dimensional data, Principal Component Analysis (PCA) is applied to streamline data dimensions. In order to address class imbalance, the SMOTE technique has been employed for oversampling the minority class of the dataset. The model's robustness is validated through evaluation on the well-established 'DataCo smart supply chain for big data analysis' dataset, yielding impressive results. The proposed approach achieves a 94.71% fraud detection rate and an overall accuracy of 99.42%. Comparative analysis with various other models highlights the significant improvements in fraud transaction detection achieved by this approach. While the model demonstrates high accuracy, it may not be directly transferable to more diverse or real-world datasets. As part of future work, the model can be tested on more varied datasets and refined to enhance generalizability, better aligning it with real-world scenarios. This will include addressing potential overfitting to the specific dataset used and ensuring further validation across different environments to confirm the model's robustness and generalizability.

[...] Read more.
Copyright Protection and Illegal Distributor Identification for Video-on-demand Applications using Forensic Watermarking

By Ayesha Shaik Masilamani V.

DOI: https://doi.org/10.5815/ijisa.2025.02.03, Pub. Date: 8 Apr. 2025

In the direct-to-home (DTH) environment video-on-demand (VOD) applications are tremendously popular due to its inexpensive and convenient nature.  In VOD approach legal customers can connect their set-top boxes (STB) to the Internet and can access or record the available content. Due to the easy transmission of the highest quality digital data to the customers by the pay-per-view approach, the data are highly at risk. The data can be vulnerable for illegal distribution of duplicate copies and they are prone to unnecessary modifications which creates a financial loss to the information creators. So it is necessary to authenticate the owner as well as the illegal distributor to reduce the digital piracy which is the motivation for this work. This paper presents a forensic watermarking scheme for protecting copyrights, and for identifying the illegal distributor who distributes the legal copy in the illegal fashion though it is copyright violation. In this paper, two watermarks are embedded in the video that is on-demand, where one watermark is the owner’s information and another watermark is related to the unique information of the STB.  This work is also suitable for the biomedical domain, where one watermark can be the patient information and another watermark will be the health center information, in order to secure the patient information and the hospital information.

[...] Read more.
Performance Analysis of Shallow and Deep Learning Classifiers Leveraging the CICIDS 2017 Dataset

By Edosa Osa Emmanuel J. Edifon Solomon Igori

DOI: https://doi.org/10.5815/ijisa.2025.02.04, Pub. Date: 8 Apr. 2025

In order to implement the advantages of machine learning in the cybersecurity ecosystem, various anomaly detection-based models are being developed owing to their ability to flag zero-day attacks over their signature-based counterparts. The development of these anomaly detection-based models depends heavily on the dataset being employed in terms of factors such as wide attack pool or diversity. The CICIDS 2017 stands out as a relevant dataset in this regard. This work involves an analytical comparison of the performances by selected shallow machine learning algorithms as well as a deep learning algorithm leveraging the CICIDS 2017 dataset. The dataset was imported, pre-processed and necessary feature selection and engineering carried out for the shallow learning and deep learning scenarios respectively. Outcomes from the study show that the deep learning model presented the highest performance of all with respect to accuracy score, having percentage value as high as 99.71% but took the longest time to process with 550 seconds. Furthermore, some shallow learning classifiers such as Decision Tree and Random Forest took less processing time (4.567 and 3.95 seconds respectively) but had slightly less accuracy scores than the deep learning model with the CICIDS 2017 dataset. Results from our study show that Deep Neural Network is a viable model for intrusion detection with the CICIDS 2017 dataset. Furthermore, the results of this study are to provide information that may influence choices while developing machine learning based intrusion detection systems with the CICIDS 2017 dataset.

[...] Read more.
Intelligent Application for Textual Content Authorship Identification based on Machine Learning and Sentiment Analysis

By Dmytro Uhryn Victoria Vysotska Lyubomyr Chyrun Sofia Chyrun Cennuo Hu Yuriy Ushenko

DOI: https://doi.org/10.5815/ijisa.2025.02.05, Pub. Date: 8 Apr. 2025

During the development and implementation of the software system for text analysis, attention was focused on the morphological, syntactic and stylistic levels of the language, which made it possible to develop detailed profiles of authorship for various writers. The main goal of the system is to automate the process of identifying authorship and detecting plagiarism, which ensures the protection of intellectual property and contributes to the preservation of cultural heritage. The scientific novelty of the research was manifested in the development of specific algorithms adapted to the peculiarities of the natural language, as well as in the use of advanced technologies, such as deep learning and big data. The introduction of the interdisciplinary approach, which combines computer science, linguistics, and literary studies, has opened up new perspectives for the detailed analysis of scholarly works. The results of the work confirm the high efficiency and accuracy of the system in authorship identification, which can serve as an essential tool for scientists, publishers, and law enforcement agencies. In addition to technical aspects, it is vital to take into account ethical issues related to confidentiality and copyright protection, which puts under control not only the technological side of the process but also moral and legal norms. Thus, the work revealed the importance and potential of using modern text processing methods for improving literary analysis and protecting cultural heritage, which makes it significant for further research and practical use in this area.

[...] Read more.
Enhancing Suicide Risk Prediction through BERT: Leveraging Textual Biomarkers for Early Detection

By Karan Bajaj Mukesh Kumar Shaily Jain Vivek Bhardwaj Sahil Walia

DOI: https://doi.org/10.5815/ijisa.2025.02.06, Pub. Date: 8 Apr. 2025

Suicide remains a critical global public health issue, claiming vast number of lives each year. Traditional assessment methods, often reliant on subjective evaluations, have limited effectiveness. This study examines the potential of Bidirectional Encoder Representations from Transformers (BERT) in revolutionizing suicide risk prediction by extracting textual biomarkers from relevant data. The research focuses on the efficacy of BERT in classifying suicide-related text data and introduces a novel BERT-based approach that achieves state-of-the-art accuracy, surpassing 97%. These findings highlight BERT's exceptional capability in handling complex text classification tasks, suggesting broad applicability in mental healthcare. The application of Artificial Intelligence (AI) in mental health poses unique challenges, including the absence of established biological markers for suicide risk and the dependence on subjective data, which necessitates careful consideration of potential biases in training datasets. Additionally, ethical considerations surrounding data privacy and responsible AI development are paramount. This study emphasizes the substantial potential of BERT and similar Natural Language Processing (NLP) techniques to significantly improve the accuracy and effectiveness of suicide risk prediction, paving the way for enhanced early detection and intervention strategies. The research acknowledges the inherent limitations of AI-based approaches and stresses the importance of ongoing efforts to address these issues, ensuring ethical and responsible AI application in mental health.

[...] Read more.