Journal of Artificial Intelligence and Computational Technology

Adaptive Multi-Scale Feature Extraction for Cervical Cancer Classification Using Dynamic Hierarchical Pooling

Abdalla Ibrahim Abdalla Musa — 2025-04-01

Cervical cancer remains a leading cause of mortality among women worldwide, especially in low-resource settings where access to early screening and treatment is limited. Early detection through accurate and efficient diagnostic methods is critical for improving patient outcomes. This study proposes a novel method for classifying cervical cancer using Dynamic Hierarchical Pooling (DHP). To effectively capture multi-scale characteristics, DHP adaptively modifies the number of pyramid levels and pooling types according to the size of the input image. To be more precise, the module dynamically divides the feature maps into different spatial regions and applies various pooling operations to each region. This adaptive mechanism extracts fine-grained and coarse-grained features crucial for recognizing diverse patterns in cervical pap smear images. To facilitate efficient processing, feature maps are resized to a common size, regardless of the original image size. The Squeeze-and-Excitation (SE) attention module further enhances feature discrimination by dynamically updating attention weights, focusing on the most informative regions of the feature maps. Combining the strengths of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), a hybrid architecture is employed to leverage local and global contextual information. Experimental results demonstrate the superior performance of the proposed method compared to state-of-the-art techniques, highlighting its potential for improving cervical cancer diagnosis

Hand Gesture Recognition for Sign Language Translation

Jithu Varghese Jacob — 2025-04-01

This project aims to develop software that translates Indian Sign Language (ISL) hand gestures into real-time text and speech. A custom symbol module enables users to add new gestures, enhancing user experience. Voice call integration allows real-time gesture-to-speech translation during calls. Optimized machine learning models ensure efficiency and accessibility, creating a solution for inclusive communication. The domain for hand gesture recognition in sign language translation, where the development of automated systems for the interpretation of sign languages using gestures would make fluid communication possible with deaf or hard-of-hearing patients. The recognition of hand gestures relies on the combination of algorithms in
image processing and machine learning to capture, understand, and translate signs into text or spoken language. Over time, these systems have evolved from the traditional rule-based and statistical methods to more advanced models that focus on achieving greater accuracy and reliable recognition through machine learning, especially deep learning.

AI-Driven Cybersecurity: Empirical Analysis of ChatGPT's Impact and Expert Perceptions

Sayeed Salih — 2025-04-02

This study examines ChatGPT's role in cybersecurity, emphasizing its potential, limitations, and perceptions among professionals. A survey targeting 200 cybersecurity experts across various industries was conducted using email and LinkedIn to gather data on familiarity with AI, ChatGPT usage, and expectations. The survey, structured into five sections, employed Likert scales, multiple-choice questions, and open-ended prompts to assess socio-demographic characteristics, familiarity with AI, ChatGPT's perceived capabilities, its expected impact, and associated challenges. Quantitative analysis, including chi-square tests and correlation analysis, revealed that 70% of respondents were familiar with ChatGPT, with 61% expecting it to enhance phishing detection and 57% highlighting its potential in malware analysis. Thematic analysis of qualitative responses identified preferences for using ChatGPT in automating routine tasks like incident reporting and responding to security inquiries. Key challenges included data privacy concerns (45%), integration barriers (50%), and ethical implications. Despite these barriers, 65% of respondents anticipated ChatGPT to improve efficiency in cybersecurity operations. These findings provide actionable insights into ChatGPT's integration in cybersecurity frameworks, emphasizing the need for robust policies, ethical considerations, and ongoing professional training to address adoption challenges.

Machine Learning Approach Integrated with MoSCoW Method for Parallel Requirements Priorization

Kawthar Ishag Ali Fadlallah — 2025-04-01

Requirements prioritization (RP) is one of the vital activities carried out through requirements engineering process. Requirements prioritization includes the selection of requirements that are reflected more important from elicited list of stakeholders' requirements. Making an incorrect selection will not only reduction the quality of the developed software but it will also earn extra cost for refinement processes in later stages. Thus, requirements prioritization would aid to determine the most appropriate requirements in different software product releases. Many research focusing on prioritizing the requirements using one or several criteria like time, dependency, and scalability. However, most of these studies address sequential prioritization only. To the best of our knowledge, no research has explored parallel ranking in prioritization, which allows for simultaneous requirements implementation, thereby reducing implementation time. Furthermore, as the volume of requirements grows, scalability becomes a critical issue. Manual prioritization is time-consuming and increases the likelihood of overlooking essential. Machine learning is increasingly popular for automating requirements prioritization. In this study we developed automated parallel requirements prioritization approach (APRP) for determine the requirements priority level in parallel format using Random Forest classifier based MoSCoW method (RF-MM). The proposed approach consists of two main modules, data elicitation and pre-processing module and prioritization module, which include established weight assignment, MoSCoW parallel prioritization, and classifier methods. Experiments on the industrial dataset (Testcase MIS system with priority) revealed that the total implementation time for sequential ranking was 76.0 seconds, whereas it was reduced to 33.0 seconds for parallel ranking. Thus, parallel ranking reduced implementation time by more than half. We achieved a maximum accuracy of 94.87%, precision of 92.31%, and recall of 92.31%.

HealthInspector: A Novel Approach to Calculating Harm Scores for Food Products Based on Chemical Composition and Nutritional Profiles

Alwin Jaison — 2025-04-01

Growing Consciousness about food safety and cosmetic products formulation has resulted in the need for systems that measure probable health hazards linked to different products.HealthInspector is a web-based system using machine learning technology to analyze food and cosmetic products based on their nutritional and chemical composition. The system drives a harm score using pre-established safety thresholds for unsafe ingredients and healthy limits for good ingredients. The backend, implemented in FastAPI, handles user input and fetched data from a PostgreSQL database, while the React.js fronted present an easy to use interface to compare harm scores of various products. The scoring model uses a rule-based penalty -reword scheme, and future versions will utilize machine learning model like XGBoost and lightGBM for improved accuracy and predictive power. This study emphasizes the value of ingredients disclosure and nutritional consciousness, empowering customers with live harm score analysis.The system is scalable, flexible, and able to adapt with emerging scientific knowledge on food and cosmetic safety.

Indian Sign Language Detection and Translation Using Deep Learning

Jishnu Plavinchottil Jayaraj — 2025-04-03

Out of India's population, about 63 million use Indian Sign Language (ISL) as the natural means of communication. However, massive barriers in communication exist between hearing-impaired people and the general population, mainly in spheres like education, healthcare, and jurisprudence, which often require professional interpreters. This language gap brings before the community of hearing-impaired several social, academic, and professional issues. The recent progress on deep learning, especially the models and architectures based on Convolutional Neural Networks (CNNs) and Transformers, have demonstrated promising results in sign language recognition. These models can be employed for significant accuracy, robustness, and better use in communication gap bridging. The project aims to develop and optimize deep learning-based sign language recognition models using the INCLUDE dataset, the standardized resource for ISL gestures. A systematic comparison and evaluation on the performance of different models will be performed on exactly the same set of data. This research therefore contributes to work on sign language recognition, pointing toward possible future solutions for a real-time translation facility and communication systems for hearing-impaired people via accurate recognition of ISL gestures. In the end, it aims at improving accessibility and promoting inclusivity in a society where communication barriers still exist for the hearing-impaired.