Journal of Information Systems Research and Practice

Preface

2025-01-06T07:06:00+08:00

The Volume 2 Issue 5 of JISRP: SUPPLEMENTARY ISSUE on Soft Computing and Data Mining (SCDM) availables online on December 14th, 2024.
The issue has been available online for accepted papers (uncorrected proofs).

Detection and Tracking of People in a Dense Crowd through Deep Learning Approach: A Systematic Literature Review

2025-01-06T07:17:04+08:00

Crowd-related incidents, such as the Hillsborough Disaster and the Kanjuruhan Stadium stampede, often result from poor crowd management, leading to tragedies like suffocation and crushing. To mitigate human error in crowd control, this research explores the use of deep learning for the detection and tracking of individuals in dense crowds. The study focuses on implementing artificial intelligence for automated crowd monitoring through a localization map, with an emphasis on re-identification accuracy and auto-annotation of targets in datasets. A systematic literature review (SLR) was conducted following the PRISMA guidelines, analyzing 4384 articles published between 2019 and 2024 across five databases. 13 primary studies met the inclusion criteria and were analyzed to address questions related to the accuracy of crowd tracking and detection. This SLR aims to provide insights and reference points for further research in artificial intelligence, particularly in the areas of auto annotation and re-identification for crowd monitoring.

Hyperparameter Tuning Modeling for Socioeconomic based Academic Analysis

2025-01-06T07:38:00+08:00

This study analyzes the impact of hyperparameter tuning in improving the performance of predictive models for academic success based on socioeconomic data. To analyze their predictive capabilities, this research focuses on two ML algorithms, Gradient Boosting Machine (GBM) and Random Forest (RF). Using a UCI Machine Learning Repository dataset, this study implements grid search for hyperparameter tuning, optimizing parameters such as learning speed and number of estimators. Results show that GBM consistently outperforms RF, with higher average accuracy (78.64% vs. 77.45%), lower standard deviation (0.0077 vs. 0.0091), and better stability. Paired t-test results (p-value = 0.0081) confirmed the statistical significance of the superiority of GBM. This research contributes to the field by integrating socioeconomic factors into academic performance prediction models, providing valuable insights for data-driven educational decision-making.

Chaos-Driven Encryption: A New Frontier For IoT Image Security Using Multi-Linear Systems

2025-01-06T07:41:59+08:00

In the context of an increasingly digital world, safeguarding sensitive visual information from unauthorized access is essential, particularly within resource-constrained Internet of Things (IoT) environments. This study introduces a novel image encryption method leveraging a Pseudo Hadamard Transformation (PHT), designed to provide a lightweight and efficient alternative to conventional pixel scrambling techniques. The proposed approach integrates chaos-based diffusion methods, which significantly enhance the encryption framework by effectively obscuring the correlation between the original image (plaintext) and the encrypted image (ciphertext). Through rigorous evaluations, the method demonstrates impressive statistical security metrics, achieving a Number of Pixels Change Rate (NPCR) of 99.6064 for the Lenna image, indicating a high degree of pixel alteration in response to single-pixel changes. Additionally, a Unified Average Changing Intensity (UACI) value of 33.4682 for the Lenna image highlights considerable intensity variations, further reinforcing the encryption's robustness. Compared to existing encryption techniques, the proposed method excels in both NPCR and UACI values, underscoring its superior performance and security capabilities. This hybrid encryption scheme, characterized by its efficient computational requirements and strong security features, is particularly well-suited for IoT applications, where maintaining a balance between data protection and resource limitations is paramount. The findings suggest that PHT, coupled with chaos-based diffusion, offers a promising solution for enhancing visual data security in modern digital environments.

Performance Evaluation using IPL Performance Impact Model

2025-01-06T07:50:57+08:00

In the context of the Indian Premier League (IPL), assessing player performance is crucial for team success and strategic planning, as the tournament demands players who can balance high-scoring rates with consistency and reliability. Player performance evaluations help teams identify top-performing individuals who contribute significantly to both offense and defense, supporting optimal team compositions. This study delves into evaluating IPL 2024 player performances through a machine learning-driven model, designed to calculate impact scores that reveal each player’s contribution across batting, bowling, and all-rounder roles. By integrating Euclidean and perpendicular distances from origin-referenced metrics, the model identifies players with a balanced performance profile across key indicators like strike rate, batting and bowling averages, and consistency. This data-driven analysis helps identify potential retention candidates for the IPL 2025 Mega Auction, offering IPL teams objective insights into players who bring strategic value and performance reliability, thus optimizing team compositions for future tournaments.

An Improved Detection and Diagnosis Methods for Breast Tumor Detection System using IEMD Transform-based CNN Architecture

2025-01-06T08:08:47+08:00

Mammogram is used to screen the tumor pixels in the breast portion of the human body. This paper develops a methodology for detecting and diagnosing the tumor regions in mammogram images using deep learning algorithm. The methodology consists of data augmentation, transform, and Convolutional Neural Networks (CNN) classifier with segmentation process for detecting the malignant mammogram image from the benign mammogram image. Then, the malignant mammogram image is analyzed for its severity levels using CNN classifier. The proposed system diagnosed the malignant case mammogram images into either Early or Advance using the developed CNN architecture. The simulation results of the proposed methods are compared with respect to other similar studies in this field.

ChatGPT's Healing Words: Navigating the Frontiers of Conversational AI in Healthcare

2025-01-06T08:13:18+08:00

This comprehensive review paper investigates the unique role of ChatGPT in medical healthcare, exploring its effect on patient engagement, clinical support, and healthcare communication by dissecting the strengths and weaknesses of ChatGPT in this domain. It highlights the potential to enhance patient care, streamlining administrative tasks to foster a more accessible healthcare experience. The paper navigates through real-world applications, shedding light on successful implementations, challenges, and potential ethical considerations. Additionally, we examine the evolving frontiers of conversational AI in healthcare, providing insights into prospects and avenues for further research. This exploration aims to showcase the current state of ChatGPT in healthcare and guide future developments in using conversational AI to improve patient outcomes and overall healthcare delivery. Further, to fully harness the benefits while minimizing the risks of ChatGPT in healthcare, it is essential for policymakers, healthcare organizations, and researchers to carefully evaluate the necessary ethical and regulatory frameworks. This paper serves as a foundation for future research and development by providing valuable insights into the advantages, disadvantages, opportunities, and risks of integrating ChatGPT into healthcare. By effectively utilizing conversational AI, we can improve patient outcomes and enhance overall healthcare services by improving patient care, expediting administrative duties and promoting accessibility.

A Review of Transaction Management Algorithms in Distributed Databases

2025-01-06T09:18:14+08:00

This paper provides a comprehensive review of transaction management algorithms in distributed databases, which are a critical part of modern IT infrastructure. The paper discusses traditional protocols such as Two-Phase Commit (2PC) and Three-Phase Commit (3PC), exploring their limitations in terms of scalability, fault tolerance, and concurrency control. Additionally, new decentralized approaches leveraging blockchain technology and smart contracts are examined, with a focus on how Proof of Work (PoW) and Proof of Stake (PoS) are utilized to enhance transaction transparency and security. Despite significant advancements, challenges remain, particularly in achieving strong concurrency control and ensuring fault tolerance among nodes. The paper concludes with a discussion of emerging trends and the potential of blockchain technology to address these issues, offering a transformative perspective on the management of distributed transactions.

Depth Examination of Deep Learning Applications in Crowd Analysis

2025-01-06T09:28:54+08:00

Artificial neural networks and machine learning have a long history of tackling diverse problems. However, as problems grew in complexity and datasets expanded, computational demands increased significantly. This paper provides a concise overview of the evolution from artificial neural networks to deep learning, highlighting models, applications, and essential technical details, including hardware, software, and libraries. It particularly focuses on conventional methods employed in crowd analysis and delves deeply into the applications of deep learning in this field, accompanied by an examination of relevant datasets. Additionally, recent studies in crowd analysis are meticulously analyzed and compared. In conclusion, crowd analysis stands as both an academic pursuit and a practical domain where deep learning has led to remarkable achievements.

Internet of Things Performance using Raspberry Pi for Body Temperature and ECG with GPS

2025-01-06T09:22:40+08:00

This paper aims to discuss how the research produced unique results that are comparable to the results of traditional devices for ECG and blood temperature, as well as the patient's location. It does this by examining previous studies, the study's criteria, and the proposed results. The majority of earlier research findings made it impossible to quickly and effectively compare patient data. Since all sensors must be used to show general results, the majority of earlier studies were unable to display results when only one sensor was used. The results of this study show that the most expensive traditional devices can be replaced with sensors and technologies that connect to the Internet of Things and Raspberry Pi to accomplish similar tasks with more flexibility when handling patient data. Aside from financial savings, integrating multiple sensors into a single device is a unique and innovative process that saves a much of time and effort. When handling his medical cases, the doctor can have more privacy thanks to the desktop system's data security, which also keeps unwanted parties from accessing patient information.

Feathers and Pixels: A Comparative Analysis of CNN Models for Efficient Bird Species Identification

2025-01-06T04:32:48+08:00

Bird species identification is crucial for biodiversity conservation, ecosystem monitoring, and avian research. It helps assess population trends, monitor ecosystem health, and track environmental changes. Birds are key indicators of habitat quality and climate impacts. Accurate identification supports studies on migration, breeding, and foraging, aiding conservation efforts and identifying threatened species. Deep learning technology, specifically Convolutional Neural Networks (CNNs), is used for bird species identification by training models on large labeled datasets of bird images. The CNN architecture is designed to extract features from images, and the model is trained to associate these features with bird species labels. The paper navigates through the nuances of model performance, facilitating a comprehensive understanding of their strengths and limitations. Metrics such as accuracy, recall, and precision are scrutinized, offering valuable insights for researchers and practitioners. The few models that were picked out for performance analysis are VGG16, VGG19, ResNet50, DenseNet, InceptionV3, and EfficientNet. The analysis was performed on 200 sample images, each of 50 different species, taken from Google’s dataset. This comparative study welcomes the reader to the intersection of avian ecology and deep learning, where feathers meet pixels to unlock new dimensions in bird species identification.