Large language models (LLMs) often underperform in low-resource linguistic environments due to under-representation in pre-training data. This work presents NaijaLLaMA-8B, a parameter-efficient adaptation of Meta’s LLaMA-3.1 8B model for Nigerian English and Nigerian Pidgin using the QLoRA fine-tuningapproach. A subset of 7,000 samples was extracted from the NaijaWeb corpus and used to perform supervised fine-tuning on a single Tesla T4 GPU under strict resource constraints. Performance evaluation shows consistent improvements over the base model, with training loss decreasing from 2.04 to 1.98 and perplexity reducing from 8.20 to 7.38. Small but measurable gains were also observed in BLEU, ROUGE-L, and BERTScore-F1 metrics. Although absolute improvements remain modest, the results validate the technicalfeasibility of adapting large language models to Nigerian linguistic contexts using limited compute and dataset size. This study establishes a reproducible baseline for Nigerian-focused language model adaptation and demonstrates the practical viability of parameter-efficient fine-tuning under constrained computational environments Received 31 March 2026Accepted 20 May 2026Published 23 June 2026
This research addresses the critical challenge of detecting fraudulent Ethereum transactions, which remains notoriously difficult due to the overwhelming prevalence of legitimate activities compared to illicit ones. This pronounced class disparity frequently results in skewed outcomes when employing conventional detection frameworks. Our investigation implements a composite resampling strategy, SMOTE-Tomek, which concurrently augments the under-represented fraud category whilst eliminating ambiguous instances from the dominant category. Subsequently, the equalised dataset undergoes processing via a stacked ensemble framework comprising three robust gradient boosting methodologies: XGBoost, LightGBM, and CatBoost, synthesised by a logistic regression meta-classifier. Through stratified 5-fold cross-validation conducted against established benchmarks, the proposed ensemble attained a mean accuracy of 0.9879, whilst precision, recall, and F1-score each achieved an exceptional score of 0.98. Benchmark comparisons verified that our stacked methodology consistently surpasses individual base algorithms. The study demonstrated that combining targeted resampling with ensemble stacking offers highly effective solutions for Ethereum fraud detection, though real-time implementation and cross-blockchain generalisability require further investigation. The proposed model provides cryptocurrency exchanges, investors, and regulatory bodies with a robust mechanism for identifying fraudulent activities, thereby enhancing ecosystem security and maintaining user confidence. This work represents the first comprehensive integration of SMOTE-Tomek resampling with stacked gradient boosting ensembles for Ethereum fraud detection, validated through rigorous k-fold cross-validation rather than simple train-test splits. Received 29 December 2025Accepted 02 February 2026Published 26 April 2026
This research addresses the challenges of energy administration and security in developing nations. It proposes an Internet of Things (IoT)-based smart meter monitoring system that integrates security controls to prevent theft and allows for remote monitoring of energy usage. The system utilizes an Atmega microcontroller for connectivity, a GSM module for sending data, a non-invasive AC current sensor for accurate readings, and a user interface for data visualization and control. Upon detecting theft, the system sends an alert to the user and disconnects the power supply. Users can also remotely monitor energy consumption and turn off the meter through the GSM connection.Received 12 February 2026Accepted 21 March 2026Published 10 June 2026
Social media have become a primary source of information across the world, shaping public discourse, opinion formation, and everyday decision-making in real time. The primary languages for discourse in the social media platforms are either, high-resource languages or low-resource. However, existing hate speech detection research has primarily focused on high-resource languages, highlighting a significant gap and need in developing detection system tailored for low-resource languages in general and the Kiswahili content in particular. The situation is exacerbated by the fact that, hate speech within social media is increasingly expressed through multimodal communication, shifting from purely text-based insults to more complex forms in which harmful intent is conveyed through combinations of text, images, videos, audio, and other rich media. To address the exiting gap, we created ChujaHate, integrating socio-cultural qualitative analysis with supervised ML for Kiswahili text-image hate speech detection. A mixed method was utilized where qualitative methods incorporated socio-cultural insights into understanding hate speech, while quantitative methods facilitated model development, evaluation, and generalisation of findings. Over 115,204 publicly available Kiswahili text sample were extracted from X (formerly Twitter) and Facebook and annotated into nine classes (disability, gender, race, religion, sexual, tribe, chronic disease, not hate, offensive), and 2,607 images were also annotated as hate or non-hate. Text data were pre-processed and represented using TF-IDF and Word2Vec.During multimodal integration, late fusion was performed at the decision level by combining the unimodal SoftMax probability distributions produced by the text and image models to generate the final multimodal hate-speech prediction model. The best-performing text classification model was the Bidirectional LSTM, which achieved an F-score of 97.32%, while the ResNet50V2 model for the image model attained an F-score of 89.77%. Furthermore, the multimodal fusion model that integrated both text (Bidirectional LSTM) and image (ResNet50V2) modalities achieved an F-score of 92.11% (precision: 91.87% recall: 92.35%) demonstrating the effectiveness of combining modalities for improved detection. This advances Natural Language Processing (NLP) by introducing ChujaHate model, with implications for moderating hate speech content, supporting safer digital spaces, and reducing online harm for the Kiswahili community. Received 10 April 2026Accepted 01 June 2026Published 08 June 2026
Cuckoo Search Algorithm (CSA) is one of the metaheuristic approaches applied by existing studies in test case optimization due to its simplicity and ease of implementation. However, the current studies lack systematic documentation and empirical validation of parameters thus limiting optimization efficiency and reproducibility. To solve this gap, the study introduces expert consensus on vital parameters for designing the Enhanced Adaptive Cuckoo Search Technique (EACST) through a planned Parameter Selection and Validation Framework. The study conducted an expert opinion survey involving 56 software testing professionals, where parameters were identified through an extensive literature review and polished using expert responses. These responses were analyzed using the Fuzzy Delphi Method which incorporates a consensus thresholds and defuzzification to assess agreement and rank parameters. The findings indicate that all parameters identified satisfy the threshold condition and attained consensus levels above 75%, with an overall agreement of 91%, indicating strong expert consensus. All parameters were accepted (α-cut ≥ 0.5), demonstrating their suitability for inclusion in the proposed technique. These results provide a structured and reproducible framework for selecting parameters, which supports an enhanced balance between exploration and exploitation in test case optimization. Nevertheless, the study is constrained to expert-based evaluation and needs further experimental validation through implementation and evaluation. The framework is worth of configuring adaptive metaheuristic techniques and represents a new integration of literature-based parameter selection with Fuzzy Delphi analysis, bridging the gap between theoretical identification and practical implementation. Received 09 April 2026Accepted 01 June 2026Published 07 June 2026
Purpose. The authors, funded by MOST-BARMM conducted a study in 2024 to monitor the surrounding environment of the headquarters of the Ministry of Science and Technology – BARMM in Cotabato City, Philippines. Relevant to this purpose, a discreet understanding on the dynamics of the weather condition coupled with the contextual existence of the MOST building in that area is crucial.Approach. This study utilized IoT-Knowledge Based Architecture in developing the automation with Arduino technology for its programming. They established three station locations: MOST Building Station (entrance); MOST 1 Station (rightmost); and MOST Lab Station (back of building). A third-party Semaphore SMS Gateway (SMS Gateway Philippines - SMS API | Semaphore (n.d.)) was utilized for massive alert reporting to system-registered Mobile Phone numbers. They leveraged the Semaphore database as the working memory stipulated (Amit Konar, 2000). The developed gateway database is for browsing operation of the system where MOST intends to retrieve data for a certain period for enhanced knowledge-based management strategy.Findings. After the development of the system the authors found out that the null hypothesis of this study “there are no significant differences between firing of a rule at particular time and firing multiple-rule simultaneously at a particular time in the knowledge-based system’s inference engine as far as Latency; and Accuracy and coherence of output directives/advices/device activations for BARMM compound constituents by the system’s intelligence or KBS” is accepted.Research implications. Firstly, the output of this study is a good reference for neophyte AI researchers from BARMM in area automation leveraging internet of things technology (IoT). Secondly, the MOST-BARMM can very well scale-up the architecture being utilized by the authors and implement it region-wide. It can monitor the environment and weather condition of the whole autonomous region by province, cities, and municipalities cost-effectively real-time and act accordingly. Practical implications. The prototype of this study effectively showed that the marriage of artificial intelligence (AI) and internet of things (IoT) would bear tremendous strategic advantage for the users especially, the MOST-BARMM employees, top to bottom not to mention the political leaders. This intelligent system would continue to inform them through their registered mobile numbers in a form of updates in relation to temperature, relative humidity, UV level, ambient light, CO2 level, vibration, rain occurrence, among other detection real-time during the day at the compound. Data extracted through the database can be analyzed for more knowledge-base management strategies.Originality. This study “Applied Intelligent Security Architecture on Continuous Detection of Environmental Data” is an original research conducted by the authors funded by the Ministry of Science and Technology within the Bangsamoro regional headquarters compound in Cotabato City. This study is anchored with the concept of artificial intelligence (AI) and the basic architecture of internet of things (IoT). It utilized the knowledge base architecture structure for the implementation of its system intelligence in software development (Amit Konar, 2000). Received 23 November 2025Accepted 12 February 2026Published 06 April 2026
The proliferation of Internet of Things (IoT) devices has created unprecedented security challenges, with traditional single-modal anomaly detection systems proving inadequate against sophisticated multi-vector attacks. This paper presents a novel cross-modal feature learning framework that synergistically integrates environmental sensor telemetry with network traffic patterns for enhanced anomaly detection across heterogeneous IoT architectures. Through comprehensive experimentation on the TON_IoT dataset encompassing 380,609 synchronized records across three distinct IoT systems (weather monitoring, smart refrigeration, and GPS tracking), we demonstrate that cross-modal integration consistently outperforms single-modal approaches. Our Random Forest implementation achieved 95.35% accuracy for weather systems (0.50% improvement over sensor-only), 78.13% for refrigeration systems (37.34% improvement), and 95.24% for GPS systems (9.45% improvement). Feature importance analysis reveals system-specific optimization patterns: atmospheric pressure emerges as the primary discriminator in weather systems (19.8% importance), while network features dominate refrigeration systems (86.6% combined importance). Most significantly, we provide the quantitative evidence that 24.7% of anomalies manifest simultaneously across both sensor and network modalities, indicating sophisticated coordinated attacks that single-modal systems would partially miss. The proposed temporal alignment methodology successfully addresses heterogeneous timestamp formats and sampling rates, creating a reusable framework for cross-modal IoT security research. These findings establish cross-modal feature learning as essential for comprehensive IoT security, with practical implications for designing resilient cyber-physical systems. Received on, 28 September 2025Accepted on, 28 January 2026Published on, 02 April 2025
The common model adopted by most departments of Computer Science (CS) in universities in Kenya for teaching CS course is the traditional faculty-led and content-driven approach that denies the learners opportunity of learning from peers or industry experts or problem-based projects. At Kisii University in Kenya, we are trying a different model: the faculty of CS teamed up with the faculties form the departments of psychology and law, and industry experts to redesign the syllabus for artificial intelligence course then delivered it using a collaborative co-teaching approach that integrated various methodologies including flipped classrooms, project-based learning, problem-based learning, and traditional lecture-based approaches. This paper reports on the perceptions and experiences of students on the proposed collaborative teaching framework for ethical artificial intelligence undergraduate level 300 course in CS curriculum and how insights from these experiences can be transferred to other undergraduate CS courses. Our aim is to advance our understanding of how to promote student-centered learning, encourage collaboration between academia and industry, and across disciplines in the university for CS curriculum delivery, and advance dialogue on integration of ethics in AI syllabus. Received: 20 January 2026Accepted: 14 March 2026Published: 14 May 2026
The rapid advancement of the Internet of Things (IoT) has significantly transformed retail automation, enhancing efficiency and customer convenience. This paper presents a Smart Trolley with IoT-Based Automatic Billing and Secure Locking System, which automates product identification, billing, and security in supermarkets. The system integrates an ESP32 microcontroller, RFID technology,and a solenoid locking mechanism to streamline the shopping experience. Customers can add or remove items using the keypad, view real-time billing details, and make secure digital payments. A solenoid lock secures the trolley at the start of shopping and remains locked until successful payment is confirmed. If a customer attempts to leave without completing payment, the system prevents trolley movement, triggering an alert via a buzzer. Additionally, all transactions and shopping data are monitored through the IoT-based ThingSpeak platform, providing real-time insights for both customers and store management. The results confirm the system’s accuracy, efficiency, and reliability, significantly reducing checkout time while enhancing product security and preventing theft. Received on, 10 July 2025 Accepted on, 28 September 2025 Published on, 05 November 2025
Cyber threats in today's software-defined network infrastructures are growing at an exponential rate, thus sophisticated, adaptive security measures are required. In order to enhance detection accuracy and computational efficiency, this study introduces a Network Intrusion Detection and Prevention System (NIDPS) that is ensemble-based and designed for Software-Defined Networking (SDN). The NIDPS makes use of a hybrid feature selection technique. As part of its soft-voting ensemble architecture, the system incorporates XGBoost, Decision Tree, and Support Vector Machine, three machine learning classifiers. To tackle the difficulties of dealing with high-dimensional network traffic data, feature selection is optimised using Correlation-Based methods and Recursive Feature Elimination (RFE). Datasets obtained from simulated Denial-of-Service (DoS) attacks were used to evaluate the model, which was constructed and tested in a virtualised, emulated SDN environment using the SEED Internet Emulator. All of the model variations demonstrated near-perfect detection ability (up to 100% accuracy) in the experiments, with the fastest predictions coming from RFE-enhanced models. The system is well-suited for implementation in actual programmable network settings due to its real-time alerting and preventive features. The results of this study show that an effective and scalable method for intrusion detection in SDNs may be achieved by combining ensemble learning with intelligent feature selection. Received: 29 November 2025Accepted: 23 January 2026Published: 22 April 2026
This paper reviews the application of computer vision and artificial intelligence (AI) in enhancing breast cancer detection, exploring how deep learning models, particularly Convolutional Neural Networks (CNNs), augment traditional screening techniques. The review examines the current state of computer vision applications in breast cancer detection, emphasizing deep learning-based approaches, and discusses how CNNs are integrated into clinical workflows, the empirical evidence supporting their effectiveness, and the practical challenges involved in their clinical adoption. The methodology also includes a deep learning-based approach to classify and segment breast ultrasound images using a publicly available dataset. CNN-based systems demonstrate performance on par with or even surpassing human radiologists in specific diagnostic tasks. Studies show that MobileNetV3, a lightweight CNN, holds strong potential for integration into edge AI systems for point-of-care diagnostics, as well as in privacy-preserving frameworks such as federated learning. The MobileNetV3-based classification model demonstrated robust performance across the three diagnostic categories: normal, benign, and malignant, with an overall test set accuracy of 91.2%. Key performance metrics, including precision (benign: 0.85, malignant: 0.74, normal: 0.83), recall (benign: 0.84, malignant: 0.74, normal: 0.88), F1-score (benign: 0.85, malignant: 0.74, normal: 0.86), and accuracy (0.82), are examined to evaluate the efficacy of these AI-driven approaches. The review identifies emerging trends, such as multi-modal learning and federated learning, which aim to enhance model robustness and privacy. The integration of AI into clinical workflows holds promise for improving diagnostic accuracy and reducing healthcare disparities by expanding access to high-quality screening services. This paper contributes to a deeper understanding of how AI-driven innovations are reshaping breast cancer detection and inspires further research toward their responsible and widespread implementation. Received on, 11 May 2025 Accepted on, 18 June 2025 Published on, 02 October 2025
Air pollution remains a critical public health and environmental challenge, especially in rapidly urbanizing regions. Accurate classification of air quality levels is essential for proactive environmental management and timely health interventions. This study presents an interpretable machine learning approach for multi-class air quality classification using logistic regression. We utilize a real-world dataset comprising 23,463 records, integrating pollutant concentrations (PM2.5, PM10, NO₂, SO₂, CO), meteorological data (temperature and humidity), and demographic features (industrial zoning and population distribution). Data preprocessing includes median imputation for missing values, feature normalization, and appropriate encoding of categorical variables. To address class imbalance, class weighting is applied, and model evaluation is conducted using 5-fold stratified cross-validation. Results show that the model achieves strong overall accuracy (87%) and a macro F1-score of 0.72, with particularly high performance for the dominant “Good” and “Moderate” categories. Feature selection methods, including Pearson correlation and recursive feature elimination, highlight PM2.5 as the most influential predictor (r = 0.98 with overall AQI). The model’s transparency and computational efficiency make it suitable for real-time deployment and policy decision-making. This work contributes a robust, interpretable baseline for air quality forecasting and highlights the importance of addressing underrepresented but critical pollution categories. Future directions include real-time data integration and comparative evaluation against more complex machine learning models. Received on, 11 June 2025Accepted on, 25 July 2025Published on, 07 October 2025
Successive industrial revolutions have reshaped the way people are managed at work. Each one brought some new inventions that made us rethink how humans and machines fit together. The Fourth Industrial Revolution (4IR) that is currently underway is characterized by a combination of technologies including Artificial Intelligence, Internet of Things, Cloud Computing, Digital Twins and more. And now, Industry 5.0 is starting to shift the focus toward something more human. It is less about pure automation and more about collaboration between humans and machines. The technologies of Industry 4.0 are transforming businesses by facilitating smart and data-driven processes. However, workforce management is a key issue in healthcare despite advancements in healthcare technology. Many healthcare organizations still rely on old-fashioned systems to manage staff. Several challenges such as high turnover rates, burnout, workload imbalances, and limited real-time visibility into operations still exist and hinder healthcare delivery. The existing solutions lack scalability, adaptability, and intelligence, leading to fragmented communication and poor decision-making. This study presents the development of a Workforce Digital Twin (WDT) for healthcare professionals using the Cross-Domain Digital Twin design pattern. This solution offers a transformative step forward in managing healthcare personnel by enabling data-driven, adaptive, and intelligent workforce strategies. Received on, 11 October 2025Accepted on, 30 November 2025Published on, 21 December 2025
This research presents a robust CNN driven biometric authentication system that combines face and iris recognition through both feature-level and score-level fusion. The framework addresses key limitations of unimodal systems which includes pose variation, lighting inconsistencies and spoofing by leveraging the strengths of each biometric trait. Two parallel CNN branches extract deep features from face and iris images, which are then fused and classified. Simultaneously, similarity scores from individual classifiers are combined using a weighted average. A hybrid decision rule integrates both outputs to enhance reliability and reduce false acceptances. The model was tested on the ORL and CASIA-IrisV4 datasets under realistic conditions. It achieved a recognition accuracy of 99.65% and a 0.00% FAR, outperforming unimodal and single-fusion approaches. This confirms the system’s potential for high-security applications. Future research will explore scalability with larger datasets, inclusion of additional modalities like fingerprint and deployment on mobile or edge devices.This research presents a robust CNN driven biometric authentication system that combines face and iris recognition through both feature-level and score-level fusion. The framework addresses key limitations of unimodal systems which includes pose variation, lighting inconsistencies and spoofing by leveraging the strengths of each biometric trait. Two parallel CNN branches extract deep features from face and iris images, which are then fused and classified. Simultaneously, similarity scores from individual classifiers are combined using a weighted average. A hybrid decision rule integrates both outputs to enhance reliability and reduce false acceptances. The model was tested on the ORL and CASIA-IrisV4 datasets under realistic conditions. It achieved a recognition accuracy of 99.65% and a 0.00% FAR, outperforming unimodal and single-fusion approaches. This confirms the system’s potential for high-security applications. Future research will explore scalability with larger datasets, inclusion of additional modalities like fingerprint and deployment on mobile or edge devices. Received on, 01 June 2025Accepted on, 06 July 2025Published on, 06 October 2025
Significant research efforts has been made towards the development of machine learning models to detect hate speech worldwide. However, for Africa, which is home to over 2,000 languages with diverse linguistic dialects, there is an urgent need for inclusive natural language processing (NLP) tools tailored to the continent's linguistic diversity. More specifically, the literature review reveals that limited research has been conducted in hate speech detection in African languages, thus providing a strong justification for this study. Whereas hate speech has continued to intrigue African communities, detection has been hampered by the complexity of multiple languages, thus calling for a localised approach to solving the problem.The study adopted the PRISMA guidelines for a systematic literature review (SLR) approach, synthesising findings from research published between 2019 and 2024 focusing on machine learning detection techniques in African low-resourced languages. The study contributes to the theoretical literature reviews and the development of Natural Language Processing (NLP) for African languages by providing a comprehensive review of research gaps in the machine learning models and datasets, highlighting the importance of multiple approaches and the need for collaborative, community-based measures to address the menace of hate speech perpetrated on social media. The findings reveal that machine learning models, including SVM, BiLSTM, mBERT, and XLM-RoBERTa, show significant potential in detecting hate speech in African languages. However, their performance is often constrained by the scarcity and limitations of available datasets. These findings provide valuable insights into the current state of hate speech detection for African languages and underscore the need to develop more comprehensive machine learning models and datasets for widely spoken African languages. Received on, 30 July 2025Accepted on, 18 October 2025 Published on, 26 November 2025
This study examines the effectiveness of swarm intelligence algorithms for optimizing MobileNet hyperparameters in breast cancer classification using ultrasound images (BCMID). Three optimization methods—Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), and the Whale Optimization Algorithm (WOA)—were applied to identify optimal learning rates, dropout rate, and the optimizer. The best hyperparameter sets discovered by each algorithm were used to retrain MobileNet to verify consistency and performance stability. The dataset consisted of clinically annotated breast ultrasound images representing benign, malignant, and normal cases. Model performance was assessed using accuracy, macro-precision, macro-recall, and macro-F1-score. The optimized models outperformed the baseline configuration, with ABC achieving 62%, PSO achieving 66%, and WOA achieving 62%. In terms of computational time, PSO required 7710 seconds, ABC 14,148 seconds, and WOA 7622 seconds, highlighting notable differences in optimization efficiency. These findings demonstrate that swarm-based optimization can enhance MobileNet’s diagnostic performance while exhibiting varying computational costs, offering a reliable framework for computer-aided breast cancer detection in ultrasound imaging. Received on, 15 November 2025 Accepted on, 24 November 2025 Published on, 22 December 2025
The development of artificial intelligence for Alzheimer’s disease (AD) diagnostics is often hindered by data privacy regulations that prevent the aggregation of sensitive patient information. Federated Learning (FL) offers a decentralized solution, enabling collaborative model training without sharing raw data. This paper presents a robust FL framework for the early detection of AD using spontaneous speech from the ADReSS dataset. We systematically evaluate the optimal components for a privacy-preserving pipeline by simulating a cross-silo federated environment. Our methodology involves comparing multiple feature extraction techniques, where VGGish audio embeddings proved most effective, and two classification models, with the Multi-Layer Perceptron (MLP) demonstrating superior performance. We further optimized the framework by comparing FedAvg, FedAvgM, and FedProx aggregation strategies, identifying FedAvgM as the most stable and effective. Our results show that the collaborative FL model significantly outperforms models trained on isolated local data. The final optimized framework achieved a state-of-the-art accuracy of 87.50% and 81.25% in a 2-client and 3-client setting, respectively. This study validates the feasibility of using federated learning to build scalable, accurate, and ethical diagnostic tools for Alzheimer’s disease. Received on, 09 August 2025 Accepted on, 25 August 2025 Published on, 11 November 2025
The purpose of this study was to develop a framework that employs multi-level access to enhance authentication mechanisms in public sector organisations and to restrict former employees whose prior elevated privileges could pose significant security risks if not effectively revoked. To achieve this purpose, the study identified multi-level policies used to control authentication in databases, examined the extent to which these policies can strengthen authentication in organisations, and developed an SQL procedure for multi-level security access authentication in database systems. Findings from the first objective revealed that most organisations have multi-level security access policies in place, which are primarily applied to database authentication based on the three security triads: confidentiality, integrity, and availability. Building on these policies, the second objective proposed a framework designed to enhance authentication and mitigate the risks posed by former employees. The framework defines three levels of authentication and recommends their implementation in public sector organisations. For the third objective, the framework was evaluated by database administration experts using SQL procedures developed from the model, and results confirmed its effectiveness in addressing the problem of access control for former employees. The study recommends the adoption of the proposed three-level security architecture, in which protection begins at the portal (level 1), continues through the engine (level 2), and extends to the database (level 3). Received on, 21 September 2025Accepted on, 23 November 2025Published on, 30 November 2025
With rapid technological progress, modern cities increasingly demand real-time responses to ensure convenience, cost-efficiency, and public safety. The IoT-Based Smart City Management project addresses this need by integrating smart technologies with civil infrastructure through a model developed in Cisco Packet Tracer, supported by Arduino Uno and relevant modules. The system combines intelligent traffic control, air pollution regulation, automated fire response, IP-based communication, and smart parking. Motion detectors enable adaptive traffic signals and real-time parking availability updates via LEDs and banners, while a pollution sensor activates a blower and CO₂ purifier to maintain air quality. Fire detection systems instantly trigger alarms and extinguishers, enhancing safety response. IP phones replace traditional telephony to reduce costs and delays. These systems function collaboratively to streamline urban management, reduce manual intervention, and enhance service delivery across multiple domains. By enabling automation and continuous monitoring, the framework not only improves urban infrastructure responsiveness but also contributes to long-term sustainability goals. As such implementations may vary by context, a trial-and-error approach is advised before wide-scale adoption. Received on, 07 October 2025Accepted on, 10 Novamber 2025 Published on, 18 Novamber 2025
Artificial intelligence (AI) technologies have advanced to the point where they can help human resource specialists, such as recruiters, by automating major parts of the hiring process and filtering the list of candidates. However, little research has evaluated the use of AI in virtual interviews. This paper presents InstaJob, an AI-powered framework designed to improve efficiency and fairness in the hiring process. It uses deep learning models for face emotion detection, text emotion analysis, and filler word detection in interviews to evaluate candidates’ soft skills, ensuring unbiased assessments. The proposed face emotion detection model achieved a validation accuracy of 77%, which outperforms the other state-of-the-art approaches.Received on, 27 April 2025Accepted on, 25 May 2025Published on, 18 June 2025
