International Journal of Emerging Multidisciplinaries: Computer Science & Artificial Intelligence

EasyGPT: A Streamlined Deep Learning Simulator

2025-04-27T10:52:49+00:00

This series of papers introduces EasyGPT, a minimalistic, flexible and novel deep learning implementation of the Transformer architecture for the simulation and testing of Natural Language Processing (NLP) applications. Built to open industry standards, our model combines a customizable modular design which enables, among other things, model selection, hyperparameter configuration and user-selectable tokenization engine plugins. In this first paper in the series, we discuss the overall system design of EasyGPT, and evaluate its performance by fine-tuning the DistilGPT2 model on the DailyDialog dataset. Our work provides both a simple way for those starting in AI research to experience ChatGTP-like chatbot technologies at the coding level, as well as providing a foundation for the transition towards more realistic and complex model-building and experimentation.

Secure File Storage on Cloud Using Hybrid Cryptography

2025-01-07T14:26:20+00:00

For decades, cloud computing has empowered the wide spread and storage of information. Its evolution has provisioned concepts for ubiquitous computing enabling accessibility to individual records without barriers to location. However, the proliferation of cloud computing provision a forum for cybercriminal to experiment. Lately, cyber-crimes proliferation has been of great concern to researchers. This have given this study the impetus to evaluate the performance of the Chacha20 and ECC algorithm while hybridizing it with added layer of security. The performance of the algorithms was evaluated against some metric including the file size and encryption and decryption time. The implemented algorithms were further compared with some of the state-of-the-art algorithm. The comparison shows that the implemented ECC and Chacha20 algorithm performed better compared to some of the compared state of the art algorithm.

Accessing Liver Disease Severity Levels from Electronic Health Records Using a Kernel-Driven Meta-Heuristic Approach

2025-04-23T07:25:18+00:00

Liver diseases are one of the major health burdens globally, affecting millions each year, with an increasing need for timely and accurate stratification of patients into various care pathways to optimize both outcomes and resources. This work uses machine learning techniques in the development of a robust model to classify liver disease patients as either inpatients or outpatients using data extracted from EHRs. The major steps involved in the process are normalization of data for feature consistency and a PCA-driven feature selection process for computational efficiency. Among the different models compared, KELM performed the best on all metrics of accuracy, precision, recall, and F1-score, closely followed by KFDA. These results emphasize the impact of preprocessing and dimensionality reduction in enhancing kernel-based algorithms and demonstrate the role of ML in clinical decision support. The approach developed is scalable, interpretable, and effective for the triage of liver disease patients and will contribute to better resource utilization and improved patient outcomes in clinical settings.

Design And Evaluation of A Language-Specific Keyboard Layout For Kanuri Language With Non-Standard Character

2025-09-13T08:58:39+00:00

The Kanuri language, spoken by over 5 million people in West Africa, faces significant challenges in digital communication due to its non-standard characters and diacritics. This study designs, evaluates, and statistically validates a linguistically optimized Kanuri keyboard layout that outperforms QWERTY by 27.7 % in speed and 33.3 % in accuracy. Applying Fitts’ Law and character frequency analysis, the proposed layout prioritizes high-frequency characters (e.g., a, n, ə) on the home row and integrates dead keys for diacritics. Usability testing with 20 native speakers demonstrated a 27.7% improvement in typing speed (23.3 WPM vs. 18.1 WPM on QWERTY) and a 33.3% reduction in error rates. Iterative refinements based on user feedback resolved key accessibility issues, such as relocating the g key and optimizing modifier layers. The findings highlight the importance of language-specific keyboard designs in promoting digital inclusion and preserving linguistic diversity. This research contributes to human-computer interaction (HCI) by providing a replicable framework for marginalized languages with non-standard orthographies.

A Genetic Algorithm-Driven Personalized Genome Mutation Pathway Predictor for Early Diagnosis of Rare Polygenic Disorders

2025-09-17T09:32:19+00:00

Accurate prediction of rare polygenic disorders remains a significant chal-lenge in precision medicine, primarily due to the fact that they involve a complicated genetic architecture and current computational models are re-stricted. Traditional polygenic risk scores (PRS) have additive assumptions and finite cross-population validity and hence are not appropriate for rare disorders. In this study, a novel GA-based approach is presented that models individualized forward mutational routes, enabling early identification of risk genomic configurations. Each GA chromosome represents a binary vector of rare variants from whole-genome sequencing data, and evolutionary processes are guided by a composite fitness function. The function integrates pathogenicity scores, disease associations, and population rarity to yield biologically relevant simulations. Using 1000 Genomes Project data, we simulate 500 mutational trajectories in 500 different individuals. Results determine an average 27.2% increase in pathogenicity and 38.4% increase in harmful variants, with more than 60% convergence to known disease profiles in European and South Asian genomes. Approximately 24% of simulated genomes per individual exceed high-risk thresholds, outperforming PRS in identifying non-additive and epistatic effects. This GA strategy offers a dynamic, ancestry-aware approach to predicting rare disease risk, broadening the scope of predictive genomics and enabling earlier, more specific clinical interventions.

Personality Prediction System via Curriculum Vitae (CV) Analysis Using Natural Language Processing (NLP) and Logistic Regression.

2025-01-10T09:41:19+00:00

When it comes to the study of humans, adjudicating one’s personality is important as it acts as a window to deciphering the individual’s mindset. The personality is a vital part of an individual when he or she works for a complex organization. There are several ways to determine an individual’s personality but the most sought after and direct method is through a simple quiz. The questions in the quiz are framed in a way that they take values with reference to the big five personality model and aid the developer in framing a personality report of the individual in question. When I take a look at the current process of hiring and selection that various organizations make use of, the employers often pick out CVs in a manual way which is monotonous, time-consuming, and consumes a lot of human resources. Our approach is rendering an automated model that motorizes the eligibility check and aptitude evaluation of an applicant in the selection process to target the drawbacks of the traditional recruitment system, a web application that analyzes both the personality and an individual’s CV has been curate. This model employs a machine learning algorithm namely “Logistic Regression” which helps to choose fair decisions to recruit a suitable candidate, and “Natural Language Processing (NLP)” uses techniques with the help of Natural Language Toolkit (NLTK) libraries to process and categorize the data. Also, the use of graphs to analyze a candidate's success makes it easier to assess his or her personality and aids in proper CV analysis. As a result, the framework lends a hand in the recruitment process, allowing the candidate's CV to be shortlisted and a reasonable decision to be reached.

Fuzzy Logic Based Personalised Diet Recommendation Engine for Dietary Prevention and Control of Diabetics

2025-03-05T13:12:15+00:00

Dietary management is a cornerstone in the prevention and treatment of Type 2 diabetes mellitus. Despite advancements in understanding dietary approaches, many patients rely on generalized advice rather than individualized plans. The affordability and accessibility of nutritious food remain significant barriers in low- and middle-income settings. Additionally, there is limited integration of technology-based tools into routine care to assist healthcare providers in delivering personalized dietary recommendations. The aim of this research is to develop a personalized food mapping system that aligns dietary recommendations with the health conditions and preferences of diabetes patients. The system derives its framework from the Nigerian food composition table thus providing culturally appropriate advice to its users. Analysing nutritional values together with local food glycemic indexes enables users to identify more suitable dietary choices which both match their nutritional requirements and food tastes. An individualised dietary system helps users maintain their planned meals more easily and these strategies work to preserve blood sugar levels. By providing recommendations that are truly individualized, it can help people better manage their blood sugar and overall health, which is essential for preventing serious long-term complications. Beyond diabetes, this personalized diet system could serve as a model for managing other health conditions. It offers a more effective way to give dietary advice, improving patient outcomes and making healthcare more efficient. Ultimately, this approach could make it easier for healthcare providers and patients alike to manage diet-related health challenges in a way that feels more personalized and adaptable.

Explainable AI in Maternal Health: Utilizing XGBoost and SHAP Values for Enhanced Risk Prediction and Interpretation

2025-02-20T11:30:53+00:00

This research investigates the integration of Explainable Artificial Intelligence (XAI) in maternal health risk prediction, with a focus on improving the transparency and clinical utility of predictive models. Maternal mortality persists as a global challenge, disproportionately affecting developing nations where healthcare systems often rely on opaque predictive tools trained on limited datasets. To address these gaps, this study analyzes a comprehensive dataset spanning clinical, physiological, and historical health metrics, applying both traditional and advanced machine learning models. By incorporating SHapley Additive exPlanations (SHAP) value analysis, the interpretability of risk predictions was enhanced while maintaining high diagnostic accuracy. The findings indicate that the XGBoost model achieved an impressive accuracy of 96.36%, with body mass index and preexisting diabetes emerging as the most significant risk determinants. Clinical insights from highly-renowned healthcare providers were actively sought during this study to contextualize the model’s implications within real-world clinical practice. These insights enable healthcare providers to prioritize high-impact variables when designing interventions, bridging the gap between algorithmic outputs and actionable clinical strategies.

When Access Fails Quietly : A Privilege Maturity and Control Drift Framework for Governance Risk in Open-Source ERP Systems

2025-08-22T07:18:51+00:00

Open-source ERP systems such as ERPNext provide flexibility for resource-constrained enterprises but often lack mature governance controls. This paper introduces a driftaware framework for access governance, centered on three novel constructs: the Privilege Maturity Index (PMI), Control Drift Taxonomy (CDT), and Access Governance Risk Score (AGRS). Validated through a longitudinal ERPNext case study in a Gulfbased firm, the model reveals how silent erosion of access discipline undermines governance integrity. Findings emphasize that systemic risks stem less from external breaches and more from organizational drift. In addition to highlighting an original framework, we show how the model naturally aligns with emerging guidance such as NIST CSF 2.0 and zero-trust architectures, ensuring both originality and applicability in modern governance contexts.

Multimodal Sensor Fusion and Adaptive Coordination Algorithms for Swarm Robotics in Disaster Response Environments.

2025-01-29T03:12:02+00:00

The increasing frequency of natural and man-made disasters highlights the urgent need for efficient response systems capable of navigating complex and hazardous environments. Swarm robotics, combined with advanced multimodal sensor fusion and adaptive coordination algorithms, offers a novel approach to addressing these challenges. This research explores the integration of diverse sensor modalities—such as thermal imaging, LiDAR, and acoustic data—into swarm robotic systems to improve real-time situational awareness and decision-making. Furthermore, we propose an adaptive coordination framework that optimizes robotic deployment, energy usage, and communication during disaster missions. Through a combination of simulations and physical experiments, the proposed system demonstrates notable advancements in victim detection accuracy, environmental mapping, and energy efficiency compared to existing methodologies. The findings of this study present a scalable and effective solution for deploying robotic swarms in disaster response scenarios, offering significant contributions to the fields of robotics and emergency management.

An Empirical Comparison of Machine Learning Algorithms for Breast Cancer Detection

2025-07-13T07:27:39+00:00

Breast cancer continues to be a major global health challenge, necessitating the development of accurate and reliable diagnostic systems. This study presents a comparative evaluation of multiple machine learning classification models aimed at enhancing breast cancer detection. Three feature selection techniques which are Principal Component Analysis (PCA), Pearson Correlation Coefficient (PCC), and Backpropagation Neural Networks (BNN) were employed to reduce dimensionality and extract relevant features. The performance of six classifiers which are Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), Decision Tree (DT), Naïve Bayes (NB), and Artificial Neural Network (ANN) was analyzed based on accuracy, precision, recall, specificity and f1-Score. Results show that among the evaluated classifiers, Random Forest and Support Vector Machine (SVM) consistently delivered the highest performance, with Random Forest achieving up to 98.8% accuracy and SVM up to 98.0%, particularly when trained on features selected through Backpropagation Neural Networks (BNN). K-Nearest Neighbors (KNN) and Artificial Neural Network (ANN) also demonstrated strong results, outperforming traditional models like Logistic Regression and Decision Tree in most scenarios. These outcomes underscore the superior classification capabilities of non-linear and ensemble-based models in handling complex feature interactions, affirming their suitability for accurate and robust breast cancer detection.

Anomaly Detection Via Network Intrusion Using A Hybrid CNN And LSTM

2025-01-09T11:09:56+00:00

The challenge posed by the measurement of quality of service (QOS) in network environments has become increasingly critical, especially in light of the constantly evolving land scape of cybersecurity threats. While various research endeavors have utilized metrices such as the false positive rate to evaluate the effectiveness of intrusion Detection systems (IDSs) in safeguarding network integrity the strategies employed by malicious actors often prioritize disrupting network and consuming resources. This problem has stimulated to undergo intensive research to develop a hybrid convolutional neural Network and Long Short-Term Memory algorithm for network anomaly intrusion detection. To evaluate the performance of the model, the precision, recall, and f1-score evaluation metrics was applied. The process of hybridization involves stacking the output of the CNN layers with the LSTM layer. The experiment revealed that network anomaly detection classifier shows a significant performance for data split of 70:30 training to testing. LSTM-CNN model was demonstrated with exceptional performance, achieving an accuracy of 99.99% after only 2 epochs. This result indicates the robustness and efficiency of the proposed hybrid mode in detecting network intrusion.

Application of Support Vector Machine for Effective Prediction of Election for Sentiment Analysis

2025-02-18T12:25:03+00:00

This study proposes the use of machine learning models, namely Support Vector Machine (SVM), for effective sentiment analysis on a dataset from the Kaggle repository. Considering the Tinubu 2023 election dataset, it can be seen that SVM having been fed with the cleansed dataset feature obtained an accuracy score of 93.2%, considering the result of each algorithm on the 2023 Nigerian election datasets. The study investigates data preprocessing techniques, feature selection, and correlation metrics to optimize the sentiment detection process. Results show that the SVM model achieves the highest accuracy, making it a potential tool for political analysis, business marketing, and public policy implementation. However, future research may explore deep learning techniques and data balancing strategies to enhance the models' performance further.

Towards Closing The Major Barrier of Adoption of Blackbox Models in The Medical Arena Based on Human-Centered XAI Design.

2025-06-10T12:16:48+00:00

The opacity of black-box models presents a significant obstacle to their acceptance in the medical field. To improve their adoption, it is crucial to identify the stakeholders who need explanations of these models and to develop effective methods for providing these explanations. This paper aims to identify the key actors/stakeholders in the medical field who require explanations of black-box models to enhance their adoption. Through a comprehensive literature review, we identify physicians, patients, regulatory bodies, ethicists, and legal professionals etc. as the primary actors with information needs regarding the workings and rationale of black-box models. Physicians require explanations to validate predictions against their clinical expertise, while patients seek transparency to understand the basis of recommendations. Regulatory bodies focus on compliance and ethical considerations, while ethicists and legal professionals evaluate fairness and accountability. By providing tailored explanations to these actors, trust can be fostered, informed decision-making facilitated, ethical concerns addressed, regulatory compliance ensured, and effective communication established. This research highlights the information needs of various stakeholders, proposes two frameworks—Human-Centered XAI Design and a workflow for black-box model research—and emphasizes the importance of explanations in enhancing the adoption of black-box models in the medical field.

Student Behaviors in College Admissions: A Survey of Agent-Based Models .

2024-12-08T08:52:50+00:00

The process of selecting colleges and securing admissions is influenced by numerous elements, particularly academic performance, behavioral tendencies, and equity considerations. Academic metrics such as high school grades and standardized exam results often form the cornerstone of admission criteria. However, behavioral factors, including decision-making styles, personal motivations, and self-image, play an equally critical role in shaping students' application choices. For instance, while some students may aspire to enroll in elite universities, others, constrained by financial limitations or self-imposed doubts, might opt for less competitive institutions. Social influences, access to advisory resources like school counselors, and awareness of the admissions process further shape students' choices and behaviors. Students from underserved or marginalized communities often face additional hurdles, leading them to prioritize institutions based on proximity, affordability, or program flexibility that aligns with their unique needs. This paper explores agent-based modeling techniques adopted by international universities to study secondary education pathways and student behaviors in the context of admissions. By examining these models, the research highlights how they simulate complex decision-making processes and systemic interactions to foster equitable practices in university admissions. Emphasizing behavioral dimensions, these models underscore the importance of creating fairer systems that address the diverse needs and aspirations of students while promoting inclusivity and justice in higher education.