LAUTECH JOURNAL OF COMPUTING AND INFORMATICS

A Quantitative and Computational Efficiency Comparison of CNN and Vision Transformer Architectures for Pneumonia Detection from Chest X-rays

Akinyemi Omololu Akinrotimi — 2026-05-11

Accurate and efficient detection of pneumonia from chest X-ray images remains a critical challenge in medical imaging, especially in resource-constrained healthcare settings. This study presents a systematic comparison between a lightweight convolutional neural network (ResNet18) and a compact Vision Transformer (ViT-tiny/16) for binary classification of pneumonia and normal cases using the publicly available Kaggle Chest X-Ray dataset. The dataset was preprocessed through resizing, normalization, augmentation, and stratified splitting into training (70%), validation (15%), and test (15%) subsets. Both models were fine-tuned from ImageNet pretrained weights and evaluated using accuracy, precision, recall, F1-score, AUROC, training time per epoch, and parameter counts. The results demonstrated that ResNet18 achieved superior recall (94.7%), F1-score (94.2%), and AUROC (0.973)
while also training faster (22.5 s/epoch) with fewer parameters (11.7M). ViT-tiny achieved marginally higher precision (94.1%) but exhibited lower recall (89.2%) and increased computational demand (35.2 s/epoch, 21.7M parameters). Interpretability analyses revealed that CNN heatmaps localized pulmonary opacities consistent with radiological patterns, while ViT attention maps distributed focus more broadly, sometimes highlighting non diagnostic regions. These findings suggest that while Vision Transformers hold promise, CNNs currently offer a more balanced trade-off between accuracy, efficiency, and interpretability in small-to-medium-scale medical imaging tasks. Future research should investigate hybrid CNN–ViT approaches, self-supervised pretraining, and multi-institutional validation to further enhance generalizability and clinical applicability.

An Ensembled Crop Recommendation System Using Soil Analysis and Image Classification

Dahiru Haruna Usman — 2026-05-11

The increasing demand for intelligent agricultural decision support systems has led to the development of numerous crop recommendation models. However, most existing systems in the literature rely solely on either soil nutrient data or image-based soil classification, resulting in limited accuracy, reduced flexibility, and poor adaptability in real-world farming scenarios, especially in resource-constrained regions. Additionally, many approaches require complete and manually inputted soil parameters, which are often unavailable to smallholder farmers. To address these limitations, this study proposes a hybrid crop recommendation system that integrates both soil nutrient analysis and image-based soil classification to improve prediction reliability and usability. The system employs a Convolutional Neural Network (CNN) for soil image classification and a Logistic Regression model for crop
prediction based on soil nutrient parameters, including Nitrogen (N), Phosphorus (P), Potassium (K), pH, and soil type. While advanced variants of CNN and more complex classifiers exist, the selected models were chosen due to their computational efficiency, interpretability, and suitability for deployment in low-resource environments. The CNN model classifies soil images into five categories Alluvial, Black, Clay, Red, and Sandy with an accuracy of 92.95%, while the Logistic Regression model achieves 87.40% accuracy in crop prediction. A hybrid decision framework is introduced to combine outputs from both models, allowing users to input either nutrient data, soil images, or both, thereby enhancing system flexibility. The system is implemented in Python and deployed using a Streamlit-based web interface, providing real-time and user-friendly crop recommendations. By integrating multiple data sources, the proposed approach improves decision accuracy, reduces dependency on complete data inputs, and supports sustainable agricultural practices. This study further highlights the need for extending crop
recommendation systems to include fertilizer type and quantity recommendations using multidimensional Hybrid Crop Recommendation System,agricultural data.

A Comparative Study of Minimalist and Maximalist UI/UX Design Approaches for Meal Plan Consultation Application

S.E. Adepoju — 2026-05-11

The evolution of user interface (UI) and user experience (UX) design continues to shape digital interaction, with minimalist and maximalist philosophies offering contrasting approaches. This study compares the usability, efficiency, and user satisfaction of minimalist and maximalist design strategies in a meal plan consultation application. A high-fidelity minimalist prototype was developed and evaluated against an existing maximalist application using a mixed-methods approach. Data from 140 participants were collected through usability testing, System Usability Scale (SUS) surveys, and qualitative feedback. Findings indicate a strong preference for the minimalist design among the study participants, with 82.1% favoring its simplicity, clarity, and visual balance. In
contrast, maximalist design was appreciated for its visual richness but was often perceived as cluttered and cognitively demanding. Younger users (18-25 years) showed the highest inclination toward minimalism; however, the sample wasb predominantly composed of younger participants, which may influence design preferences and limit the generalizability of the findings. The findings suggest that minimalist design approaches may enhance perceived usability and user experience in this context, particularly for younger and tech-savvy users. The study recommends a balanced design strategy that combines simplicity with selective expressive elements, while highlighting the need for further research using more diverse populations and controlled experimental methods.

AN ADDITIVE ABLATION STUDY ON AUTOMATED CODE GENERATION: INSIGHTS FROM ROBERTABART_X

Philip Ajibade Adedayo — 2026-05-11

Automated Code Generation (ACG) remains a trending research domain used to facilitate software development; however, previous studies have shown that ACG models often struggle with long-term context and exhibit poor domain adaptation and functional precision. As a result of this, we proposed a modular hybrid transformer, RoBERTaBART model augmented by Natural Language Processing (NLP) methods to address these limitations. We perform an ablation study on RoBERTaBART_X, a hybrid transformer that combines the context power of RoBERTa with the generation capability of BART. The following modules such as Task-Adaptive Pretraining (TAPT), Domain-Specific Data Augmentation (DA), Retrieval Augmented Generation (RAG), FlashAttention, and Sparse Attention were further utilized to enhance the hybrid model. Furthermore, the impact and the benefits of each module as well as the overall combination of the enhancements were carried out on CoNaLa, Django, and
CodeSearchNet datasets. It was shown that every module complements each other, with RoBERTaBART-X outperforming other variant models across all evaluation metrics, particularly in CodeBLEU and syntax validity. These results show that the hybrid modeling, retrieval, and iterative correction are key to significantly improving automated code generation performance.

EXPLAINABILITY FATIGUE IN ARTIFICIAL INTELLIGENCE: A PRISMA-GUIDED CONCEPTUAL FRAMEWORK OF COGNITIVE LIMITS IN HUMAN–AI INTERACTION

Dahiru Hassan Usman — 2026-05-11

Explainable Artificial Intelligence (XAI) is increasingly recognized as essential for developing responsible and trustworthy AI, predicated on the assumption that greater transparency enhances user understanding, trust, and decision-making. Despite extensive, rigors research in explainable artificial intelligence (XAI), the integration of AI into high-stakes domains remains constrained by concerns over interpretability and trust. Current technical solutions often neglect human cognitive frameworks for interpreting complex decisions, leading to a phenomenon termed “explainability fatigue” where cognitive effort required to comprehend AI explanations outweighs perceived benefits, resulting in diminished engagement and suboptimal reliance on AI systems. This research employed a PRISMA 2020-guided conceptual systematic review to synthesize theoretical and empirical work on XAI and human cognitive constraints. Following identification, screening, eligibility, and inclusion phases,
searches across major databases (IEEE Xplore, Scopus, Web of Science, ACM Digital Library, Google Scholar) yielded 32 studies from 2021-2026. Studies were systematically coded and grouped into thematic domains: cognitive load in XAI, trust calibration, interpretability techniques, and human-centered design principles. Analysis revealed that explanation complexity increases extraneous cognitive load, leading to performance degradation rather than improvement. Three key outcomes emerged: trust miscalibration (both overtrust and undertrust), degraded decision quality through cognitive overload, and accountability gaps. The synthesis identified antecedent variables (explanation complexity, volume, user characteristics, contextual constraints) that
mediate explainability fatigue. The paper proposes a framework that positions explainability fatigue as a mediating factor between explanation design and responsible AI outcomes, explainable AI systems should adopt adaptive, context-aware explanation strategies aligned with human cognitive capabilities rather than pursuing maximal transparency, marking a shift toward “cognitively sustainable transparency” in responsible AI design.

Machine Learning Driven Intrusion Detection for Internet of Things Networks: A Comparative Study of Ensemble and Traditional Models

A. Adeyemo Adetoye — 2026-05-11

This study investigates the effectiveness of an intrusion detection system (IDS) powered by machine learning (ML) for securing Internet of Things environments. The growth of IoT devices has increased exposure to complex cyber threats that traditional security mechanisms struggle to detect. In this work, a supervised learning based intrusion detection framework was developed and evaluated using the publicly available UNSW-NB15 dataset, which represents real IoT network traffic scenarios. The methodology involved data preprocessing, feature engineering, model training, and performance evaluation using multiple metrics including accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). Five ML models were examined, namely Random Forest, XGBoost, Decision Tree, Logistic Regression and Naive Bayes. Experimental results show that ensemble models achieved superior performance. Random Forest recorded the highest performance with an accuracy of 99.78%, precision of 9.59%, recall of 98.27%, F1-score of 97.93% and AUC of 99.99%, followed by XGBoost with accuracy of 99.76%, precision of 97.72%, recall of 97.84%, F1-score of 97.78% and AUC of 99.99%. Decision Tree achieved an accuracy of 99.68%, precision of 96.76%, recall of 97.28%, F1-score of 97.02% and AUC of 98.55%, while Logistic Regression recorded an accuracy of 99.28%, precision of 91.07%, recall of 95.86%, F1-score of 93.40% and AUC of 99.92%. Naive Bayes produced lower performance with accuracy of 92.91%, precision of 40.16%, recall of 66.30%, F1-score of 50.02% and AUC of 93.56%, indicating reduced capability in modeling complex IoT traffic patterns. Further analysis using confusion matrices, ROC curves, and calibration plots confirmed the robustness and reliability of ensemble approaches. The findings demonstrate that ML driven intrusion detection is effective for IoT security, with XGBoost and Random Forest offering the best balance between detection performance and false alarm reduction.

LABVIEW-BASED HUMAN-MACHINE INTERFACE FOR AN AUTOMATED INCUBATOR

Olugbenga Kayode Ogidan — 2026-05-11

Microorganisms are temperature sensitive and as such they are usually kept in temperature-controlled chambers. Many existing incubators use Liquid Crystal Display (LCD) as temperature monitors which does not give users full monitoring and control of the incubator. This research develops a Human-Machine Interface (HMI) for an automated microbial incubator to enhance user’s experience by allowing them full temperature monitoring and control without having to open the incubator from time to time. The HMI is developed with the National Instruments Laboratory Virtual Instrument Engineering Workbench (LabVIEW). Temperature acquisition and control were achieved using the Atmega 328 microcontroller on an Arduino board which was interfaced with LabVIEW. Tests performed reveal that the incubator was able to maintain a temperature between 23o C and 30o C while providing information for users through LCD, Arduino serial, and the developed LabVIEW-based user interface. The LabVIEW interface was able to provide a graphical display of incubator temperature rise and fall on a real-time basis thus enhancing users’ experience more than LCD or serial monitor. This work will add value to microbial study enabling researchers’ efficiency and productivity.

An Intelligent Dermoscopic Image Analysis Framework for Skin Cancer Detection Using Enhanced Metaheuristic-Optimised Support Vector Machines

Olusola Bamidele Ayoade (Phd) — 2026-05-11

Skin cancer is among the most common and lethal malignancies globally, with its high mortality rate primarily due
to delayed diagnosis and insufficient early detection. Despite recent advancements, current computer-aided diagnostic techniques frequently exhibit restricted classification accuracy, high computational complexity, and ineffective hyperparameter optimisation, thereby limiting their efficacy in practical clinical settings. This study aims to create efficient and robust intelligent models for precise skin cancer detection and classification by improving optimisation-driven machine learning techniques. The study proposes two enhanced multi-class Support Vector Machine (SVM)-based frameworks to achieve this objective: MCMLMBSVM-EAO and MCMLMBSVM- EHHO. The study aims to: (i) improve dermoscopic image quality via preprocessing techniques such as resizing, grayscale conversion, contrast enhancement, and noise reduction; (ii) accurately segment lesion regions using the Sobel edge detection method; (iii) extract discriminative colour, shape, and texture features using Colour Moments and Gray Level Co-occurrence Matrix (GLCM); and (iv) enhance SVM classification efficacy through sophisticated hyperparameter optimisation. To address intrinsic optimisation challenges, the Aquila Optimiser (AO) and Harris Hawks Optimiser (HHO) are methodically refined with innovative mechanisms to augment convergence speed, search efficacy, and solution robustness. The proposed models are assessed using the benchmark HAM10000 dermoscopic dataset. Experimental findings indicate that the MCMLMBSVM-EAO and MCMLMBSVM-EHHO models attain classification accuracies of 98.45% and 98.67%, respectively, surpassing numerous leading methodologies. The findings underscore the efficacy, resilience, and generalisability of the proposed frameworks, rendering them appropriate for dependable and efficient automated dermatological diagnosis, especially in resource-limited settings.

Design and Implementation of an Online Whistle-Blower Reporting System to Combat Public Property Vandalism in Lagos State

Ayorinde Felix Ajibaye — 2026-05-11

Public property vandalism presents a growing challenge in Lagos State, with existing whistle-blowing mechanisms often constrained by inadequate anonymity, weak feedback mechanisms, and limited citizen trust. This study presents the design and implementation of an Online Whistle-Blower Reporting System aimed at addressing these gaps. The system was developed using the Design Science Research Methodology (DSRM) and informed by the Technology Acceptance Model (TAM) and trust-based privacy principles. It was implemented using the MERN (MongoDB, Express.js, React.js, Node.js) stack and deployed on AWS cloud infrastructure. Key features include anonymous and confidential reporting, multimedia evidence submission, password-protected case tracking, and role-based dashboards for administrators and relevant agencies. The system was evaluated through functional testing, usability assessment, performance testing, and penetration testing, with results indicating a 93.3% functional pass rate, a 90% task completion rate for usability, and average response times of under two seconds. The system also demonstrated resilience against common vulnerabilities based on assessments using OWASP ZAP and Burp Suite. Comparative analysis with existing whistle-blowing platforms in Nigeria suggests improvements in usability, accessibility, and reporting transparency. While the system was evaluated in a controlled environment, the findings indicate its potential to support more effective reporting mechanisms and contribute to civic engagement in Lagos State.

THE ROLE OF DATA SCIENCE IN OPTIMIZING RENEWABLE ENERGY GENERATION FROM WIND FARMS: A CRITICAL CONCEPTUAL REVIEW

Dr. Dahiru Haruna Usman — 2026-05-11

This study systematically reviews the role of data science in optimizing renewable energy generation from wind farms, addressing the growing demand for efficient, reliable, and cost-effective wind energy systems in the global transition to low-carbon energy. Despite its maturity and scalability, wind energy continues to face challenges related to resource variability, forecasting uncertainty, operational inefficiencies, and high maintenance costs, necessitating advanced optimization strategies. The primary objective of this review is to examine how data-driven techniques enhance operational efficiency, system reliability, and energy output in both onshore and offshore wind farms. The study follows the PRISMA 2020 guidelines, with peer-reviewed literature sourced from Scopus, Web of Science, IEEE Xplore, and Science Direct. Studies were selected using predefined inclusion and exclusion criteria, and data were extracted on applied methods, datasets, optimization objectives, and performance metrics. A qualitative synthesis, supported by comparative performance analysis where feasible, was conducted. The findings show widespread application of machine learning, deep learning, and hybrid models, particularly in wind power forecasting, turbine performance optimization, predictive maintenance, and grid integration. Across the reviewed studies, data science techniques consistently improve forecasting accuracy, turbine efficiency, and maintenance effectiveness, resulting in reduced downtime and operational costs. Overall, data-driven approaches outperform many traditional methods in managing the complexity and variability of wind energy systems. In conclusion, the review establishes data science as a key enabler of efficient, reliable, and economically viable wind farm operations, while emphasizing the need for future research on explainable models, standardized benchmarks, and scalable, integrated optimization frameworks.

A Brain Tumor Detection And Identification System Using Large Language Model

Hanat Yetunde Raji-Lawal — 2026-05-11

Brain tumor represent a mass like structure of living and dead cells that grows uncontrollably inside the brain. Among the various medical issues, brain tumors are a big concern. They are the 10th leading cause of death in the developing world. About 700,000 people have brain tumors with 80% being non-cancerous and 20% cancerous. This study focuses on the development and evaluation of an OpenAI-based Large Language Model, for the automated classification and descriptive reporting of brain MRI images. The primary aim is to enhance diagnostic workflows by reducing human error, accelerating detection, classification and generating structured textual reports to assist radiographers.

This research employs OpenAI-based Large Language Model capabilities to process visual inputs and return structured textual outputs. This approach leverages its large-scale pre-training on aligned image-text pairs to perform semantic analysis, classification, and localization. The model was guided through structured prompt engineering to identify tumor type, size, and anatomical location. Experimental results demonstrated that the system achieved an overall classification accuracy of 90%, with recall scores of 85% for meningioma, 95% for glioma, and 90% for pituitary tumors, F1-scores across all classes ranged from approximately 0.88 to 0.94. These findings highlight the potential of OpenAI-based Large Language Model as a supportive diagnostic tool in medical
imaging.