ADAPTIVE OR STATIC IN NATURE: INVESTIGATING THE PSYCHOMETRIC QUALITY OF COMPUTER-BASED ENGLISH TEST
ALLWELL NJIGWUM (National University of Lesotho)*; OWHORCHUKWU ANWURI (Ignatius Ajuru University of Education, Port Harcourt )
Corresponding Email: allwellsn@gmail.com
In many Nigerian universities, Computer-Based Testing (CBT) has become the standard for assessing all General Studies (GNS) courses. Despite its proliferation, the unanswered question is whether the CBT systems merely shuffle fixed sets of questions (static) or adjust them to students’ abilities (adaptive). Therefore, this study conducted a psychometric evaluation of a CBT English language assessment in a Nigerian university using both Classical Test Theory (CTT) and Item Response Theory frameworks. Guided by four research questions, the study employed the instrumentation and comparative cross-sectional designs to determine the psychometric quality of the 50-item CBT English Language test administered to 550 undergraduates. Data collected was analysed using R and Jamovi statistical tools. The EFA and CFA revealed a multi-dimensional test with three dimensions. Although the test proves to be reliable, the CTT item analysis revealed that 36% (18) of the items were flawed while 64% (32) were retained. While the 2PL Multidimensional IRT analysis rejected 68% (34) of the English CBT items, retaining only 16 items. This suggests the test the CBT English test lacks the necessary psychometric soundness to function as either a valid static or adaptive CBT test. The study recommends extensive revision to strengthen construct validity and item quality of CBT assessments in Nigerian universities.
Keywords: Computer-Based Testing, Psychometric Evaluation, Item Response Theory, Classical Test Theory, Nigerian University Assessment
Assessing the Dimensionality and Model Fit of the 2015 NECO Mathematics Multiple-Choice Items in Nigeria Using the Four-Parameter Logistic Model for Achieving SDG 4
Oluwaseyi Opesemowo (University of Johannesburg, South Africa) *, Kehinde Olufunke OPATUNJI (Obafemi Awolowo University, Ile-Ife, Nigeria), Temitope BABATIMEHIN (Obafemi Awolowo University, Ile-Ife, Nigeria), Titilope Rachael OPESEMOWO (University of Johannesburg, South Africa)
Corresponding Email: oopesemowo@uj.ac.za
Education requires valid and reliable assessment systems to measure students’ learning outcomes accurately. In Nigeria, the National Examinations Council (NECO) Mathematics examination remains critical for assessing secondary school students’ mathematical proficiency. Yet, limited empirical evidence exists on the psychometric soundness of its test items using advanced Item Response Theory (IRT) models. This study investigated the dimensionality and model fit of the 2015 NECO Mathematics multiple-choice test items using the Four-Parameter Logistic (4PL) model, with a view to promoting data-driven test validation practices that advance SDG 4. An ex post facto research design was adopted. The population comprised 24,718 candidates who sat for the 2015 NECO Senior School Certificate Mathematics III Examination in Osun State, Nigeria. A multistage sampling procedure was used to select 3,000 examinees. The NECO June/July 2015 Mathematics multiple-choice paper served as the instrument, and data were analysed using the MIRT package in the R programming language. Results revealed that the test met the unidimensionality assumption of IRT, with lower AIC, SABIC, HQ, and BIC values (1,142,691; 1,144,092; 1,145,043; 1,145,043) compared to multidimensional models. The items also demonstrated good fit to the 4PL model, confirming their psychometric adequacy. The study concludes that the 2015 NECO Mathematics examination is unidimensional and effectively modelled under the 4PL framework. These findings underscore the need for continuous psychometric evaluation of national assessments to ensure fairness, accuracy, and alignment with SDG 4 objectives for quality learning and equitable educational outcomes.
Keywords: Item Response Theory (IRT); Four-Parameter Logistic Model (4PL); Model fit; Dimensionality; SDG 4.
Assessing the Psychometric Properties of Science Test: Comparing Generative AI and Traditional Statistical Tools
ALLWELL NJIGWUM (National University of Lesotho) *; Ramota Rasaq (Obafemi Awolowo University)
Corresponding Email: allwellsn@gmail.com
Generative Artificial Intelligence (GenAI) tools have proven to analyse and interpret complex educational data with a high level of precision and adaptability, making them a powerful and cheap alternative to traditional statistical programs. Though rarely explored in the field of psychometrics, this opens new avenues for educational measurement and assessment. This study compares three Generative AI platforms (Power Drill AI, Julius AI, and ChatGPT-5.0) with three conventional statistical tools (R, Stata, and jMetrik) in evaluating the psychometric properties of the Science Achievement Test. The study adopted a comparative quantitative approach under the positivist paradigm. Guided by five research questions, 1,510 Junior secondary students in Port Harcourt, Nigeria, will be drawn using multistage sampling. A standardised BECE Science test was employed for data collection. The study was anchored on the Item Response theory, a modern test framework for evaluating test quality. Data analyses were independently conducted on both Gen AI platforms and conventional statistical software under the IRT framework. The results revealed that across all parameters, ChatGPT 5.0 demonstrated comparable precision to traditional statistical software (R 4.4.1 and JMetrik) for IRT calibration. It was suggested that extensive validation studies be conducted to ratify the reliability of high-performing Gen AI models.
Keywords: Generative AI, Psychometrics, Item Response Theory, Educational Assessment, Science Achievement Test
DESIGN AND DEVELOPMENT OF AN AI-BASED ADAPTIVE ASSESSMENT SOLUTION FOR THE STUDENTS’ INDUSTRIAL WORK EXPERIENCE SCHEME IN NIGERIA
Abdulrasaq Sulyman (University of Ilorin) *; Muhammad Kamaldeen Jimoh (University of Ilorin)
Corresponding Email: sulymanabdulrasaq326@gmail.com
The Students Industrial Work Experience Scheme (SIWES) is a vital component of technical and vocational education in Nigeria, designed to bridge the gap between classroom learning and industrial practice. However, traditional paper-based evaluation methods have been plagued by inefficiencies such as falsified logbooks, delayed supervision, lack of transparency, and subjective grading. This study aimed to design and develop an AI-based adaptive assessment platform to enhance the evaluation process for students participating in SIWES. The specific objectives were to develop an adaptive platform capable of generating personalized assessments based on students’ weekly progress reports, validate its instructional, technical, and assessment quality, and compare its effectiveness with conventional SIWES evaluation methods. A research and development design guided the study, using the Interactive Waterfall Model to ensure iterative refinement and user-centred functionality. The platform was implemented using HTML, CSS, JavaScript, PHP, and MySQL. Data were collected through expert rating questionnaires and feedback forms from 10 students and 2 supervisors during the pilot phase, and 30 students and 5 supervisors during full deployment. Validation by experts in Educational Technology, Computer Science, and Measurement & Evaluation confirmed the system’s reliability, with Cronbach’s Alpha values of 0.84, 0.79, and 0.81 across instructional design, technical functionality, and assessment quality domains. Findings revealed that the developed system effectively addressed major challenges in traditional SIWES assessment. Expert ratings indicated high usability, adaptability, and efficiency (M = 3.33–3.67; SD = 0.58–1.00), while students and supervisors reported enhanced accountability, transparency, and ease of monitoring (M = 3.42; SD = 0.67). The adaptive feature successfully personalised evaluations to students’ workplace experiences. The study concludes that the AI-based platform improves fairness, transparency, and engagement in SIWES evaluation. It recommends adoption across Nigerian institutions, with further expansion using advanced AI techniques and multi-institutional validation.
Keyword: AI-based assessment, adaptive learning, SIWES evaluation, vocational education, educational technology, student supervision, Nigeria
Development of Adaptive Testing Models for Literacy Assessment in Primary Education
Sukurat Oyedokun *; Celeste Combrinck (University of Pretoria)
Corresponding Email: oyedokunolamide77@gmail.com
Literacy assessment in primary education plays a critical role in identifying learners’ strengths and addressing developmental needs to support equitable academic progress. However, traditional fixed-form literacy tests often fail to accommodate diverse learning paces, resulting in limited measurement precision and reduced student engagement. This study focuses on the development and validation of adaptive testing models designed specifically for literacy assessment in primary school populations. Grounded in Item Response Theory and advanced data analytics, the proposed Adaptive Testing on Literacy for Pupils (ATLP) dynamically adjusts item difficulty in real-time based on learner performance. A quantitative design involving stratified random sampling was employed to ensure representativeness across socio-economic and geographic contexts. Preliminary findings from pilot testing indicate that adaptive assessments enhance student engagement, improve accuracy in measuring literacy skills, and reduce test-related stress compared to conventional assessments. The integration of psychometric calibration and adaptive algorithms demonstrates strong potential for personalised and efficient literacy evaluation. The study concludes that adaptive testing models can transform assessment practices in primary education by promoting fairness, instructional responsiveness, and improved learning outcomes, while highlighting the need for continued research and infrastructural consideration to ensure successful implementation.
Keywords: Adaptive testing, Literacy assessment, Item Response Theory, Primary education
ENSURING EQUITY AND INCLUSIVITY IN AI-POWERED EDUCATIONAL ASSESSMENTS: A CASE STUDY OF POLICY AND PRACTICE IN NIGERIA
Mary Otegbade (University of the Cumberlands) *; Deborah Demurin (Lead City University); Babatunde KOLASHI (Lead City University)
Corresponding Email: otegbademary@gmail.com
Artificial Intelligence (AI) is rapidly reshaping educational assessment by enabling personalised, efficient, and scalable evaluation systems. In Nigeria, initiatives such as the Joint Admissions and Matriculation Board’s Computer-Based Testing (CBT), EdTech solutions, and digital learning programs demonstrate growing adoption of AI-driven assessments. However, persistent socio-economic divides, infrastructural limitations, and unequal access raise concerns about equity and inclusivity. Guided by the Digital Equity Framework and UNESCO’s AI in Education policy guidelines, this qualitative case study investigates how Nigerian education policies and current practices address inclusivity in AI-powered assessment. A thematic analysis of key national policy documents and implementation frameworks reveals limited provisions for marginalised learners, urban-biased deployment of digital assessment systems, weak regulatory oversight of private EdTech innovations, and insufficient attention to the needs of learners with disabilities or those requiring multilingual access. The findings highlight a misalignment between digital transformation goals and inclusive practice. The study concludes that Nigeria must strengthen policy depth, enforce ethical AI standards, and invest in inclusive digital infrastructure to ensure AI-powered assessments foster equal educational opportunities rather than deepen existing inequalities.
Keywords: AI-powered educational assessment, equity, inclusivity, Nigeria
Enhancing Test Quality in Lesotho Basic Education through Classical and Item Response Analyses
LEFA THAMAE (National University of Lesotho) *
Corresponding Email: makauclement@gmail.com
High-quality assessment is essential for improving learning outcomes, particularly in developing contexts such as Lesotho, where fairness and accuracy in testing remain challenges. Multiple-choice questions (MCQs) are common in basic education due to their efficiency and objectivity; however, poorly constructed items can compromise validity and reliability. Limited empirical evidence exists on the psychometric quality of classroom assessments in Lesotho, as teachers often rely on intuition rather than measurement models. This study developed a Grade 6 mathematics MCQ test aligned with the Cambridge International Curriculum and evaluated its psychometric properties using Classical Test Theory (CTT) and Item Response Theory (IRT). A descriptive and developmental design was employed with 200 Grade 6 learners from public and private schools. Thirty MCQ items were constructed based on a table of specifications and validated through expert review. Pilot testing supported item refinement, and JMetrik software was used for data analysis. CTT indices—including item difficulty, discrimination, and Cronbach’s alpha—were computed, while IRT analyses applied the Rasch (1PL), 2PL, and 3PL models to estimate item parameters and model fit. Findings showed moderate-to-low reliability (α = 0.62) with several items displaying weak discrimination (<0.20). The 2PL model demonstrated the best fit, indicating variability in item quality and highlighting items with extreme difficulty that contributed limited information. The study emphasises the need for rigorous item calibration to enhance fairness, diagnostic accuracy, and alignment with learner ability. It recommends institutionalising psychometric modelling and capacity building for educators to strengthen evidence-based assessment practices in Lesotho and similar contexts.
Keywords: Educational measurement, Lesotho basic education, multiple-choice test development, classical test theory, item response theory
Exploring AI Readiness and Digital Competence among Educators: Lessons for Adaptive Testing Implementation
Festus Ben (University of Johannesburg) *; Kafilah Gold (University of Johannesburg)
Corresponding Email: festobit@gmail.com
This study explores educators’ readiness to integrate Artificial Intelligence (AI) in education and its implications for adaptive testing implementation. Using aggregated insights from teacher responses, it examines levels of digital competence, familiarity with AI-supported assessment tools, and willingness to adopt adaptive testing practices. The study is guided by the Digital Competence Framework for Educators (DigCompEdu) and the Technology Acceptance Model (TAM), emphasising how professional engagement, digital resource creation, and perceived usefulness influence technology adoption. Findings show that while most teachers demonstrate moderate-to-high digital confidence, fewer feel adequately prepared to apply AI concepts to assessment design. Institutional readiness and infrastructure remain weaker than individual enthusiasm. Barriers include insufficient training, limited resources, and a lack of policy support, while key enablers are professional curiosity and the perceived usefulness of AI for assessment. The study concludes that enhancing AI readiness among educators is essential for equitable and effective adaptive testing adoption.
Keywords: Artificial Intelligence, Educators’ Readiness, Digital Competence, Adaptive Assessment, Professional Development
Harnessing Ethical AI in Student Academic Performance Prediction: A Machine Learning Perspective
Ijeoma Chikezie (National Institute for Nigerian Languages, Aba, Abia State) *; Musa Ayanwale (University of Johannesburg, Auckland Park, 2006, South Africa)
Corresponding Email: drijeomajchikezie@gmail.com
As artificial intelligence (AI) becomes increasingly embedded in educational systems, its application in student assessment and performance prediction presents both opportunities and ethical concerns. This paper explores the deployment of traditional and deep learning machine learning models to predict student academic performance using an open-source educational dataset (xAPI-Edu-Data). While the integration of AI into assessment practices can enhance precision, early intervention, and resource allocation, it also raises critical questions around fairness, explainability, and algorithmic bias. The study compares models such as Logistic Regression, Random Forest, Support Vector Machines, Artificial Neural Networks, and Long Short-Term Memory networks. Through rigorous preprocessing, training, and evaluation, the models were assessed based on accuracy, F1-score, ROC-AUC, and interpretability using SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations). The results reveal that although deep learning models provide superior accuracy, ensembling traditional models offers greater transparency and computational efficiency. Ethical implications are discussed, focusing on how demographic features (e.g., gender, nationality) influence predictions, highlighting the risk of reinforcing systemic biases. The paper underscores the importance of embedding fairness-aware mechanisms and stakeholder involvement in the deployment of AI-based assessments. It argues for the development of contextually relevant policies that uphold data privacy, equity, and explainability in line with emerging global standards. This work contributes to the assessment discourse by emphasising that while AI can augment educational evaluation, its ethical deployment must be guided by inclusive, transparent, and accountable practices. This paper concludes with recommendations for educational bodies and examination councils to adopt AI responsibly within a regulatory framework that safeguards student welfare and institutional integrity.
Keywords: Artificial Intelligence, Student Assessment, Machine Learning, Ethical Evaluation, Algorithmic Fairness
Lecturers’ Knowledge dynamics for technology integration for educational assessment in higher institutions in Nigeria
Kawu Muritala (University of Ilorin); Jumoke Oladele (University of Johannesburg) *
Corresponding Email: kawumuritala@gmail.com
This study examines the knowledge dynamics underlying the integration of technology in educational assessment within Nigerian higher institutions with the technological pedagogical content knowledge framework. A descriptive survey research design of the survey type was adopted for the study. A researcher-designed and validated questionnaire was used for data collection, and the data collected were subjected to descriptive statistics using frequency, percentage and the mean statistical techniques for answering the research questions raised and inferential statistics using Factorial Analysis of Variance to test the generated null hypothesis. Results show that lectures have a wide spectrum of technological integrations for educational assessment with high knowledge, but the lectures’ institutional type and their years of experience play important roles in their engagements, based on which the conclusion was drawn and recommendations made.
Keywords: Technology Integration, Knowledge dynamics, Educational assessment, Higher institution
Machine Learning Framework for Categorising Test Items Based on Bloom’s Taxonomy for Adaptive Testing Systems
ISAAC IFINJU (University of Ilorin) *; Mayowa Ogunjimi (University of Ilorin); Timothy Obasuyi (East Tennessee State University)
Corresponding Email: olawaleifinju@gmail.com
This study introduces a machine learning framework for automatically classifying test items based on Bloom’s taxonomy of cognitive levels. The goal is to improve adaptive testing systems. The framework is built on psychometric principles, particularly Item Response Theory (IRT), and utilises insights from previous research on cognitive assessment. It employs Natural Language Processing Techniques, specifically a transformer-based model, to analyse and categorise Economics test items. The methodology consists of several steps, starting with data preparation and feature engineering, followed by model training and evaluation. The system achieved an impressive overall accuracy of 97.7%, with precision, recall, and F1-scores all nearing 0.98, demonstrating strong performance across Bloom’s categories. Visual comparisons of actual versus predicted distributions further validated the accuracy of the classification, with only minor discrepancies noted in the “Understand” level. These results show that combining Bloom’s taxonomy with machine learning and IRT principles is possible and can improve the scalability, validity, and responsiveness of educational assessments. The study concludes that such frameworks can enhance precise item classification in adaptive testing, ensuring they align with intended learning outcomes and adapt to learner abilities.
Keywords: Machine Learning, Bloom’s Taxonomy, Adaptive Testing, NLP, Educational Data Mining, Transformers.
The Future of Lifelong Learning in the Digital Age: Implications for Assessment Practices in Nigeria
Miriam James (International Centre for Educational Evaluation (ICEE), Institute of Education, University of Ibadan) *
Corresponding Email: mjames3434@stu.ui.edu.ng
Lifelong learning has become central to 21st-century education, with digital technologies expanding opportunities for continuous knowledge and skills development in Nigeria. However, assessment practices remain dominated by traditional examinations, creating a mismatch between evolving digital learning environments and outdated evaluation methods. This study examines the future of lifelong learning in the digital age and its implications for assessment practices in Nigeria by analysing how the adoption of digital lifelong-learning platforms, access to digital infrastructure, and digital literacy influence assessment transformation. Guided by Connectivism and Transformative Learning Theory, the study employed a descriptive survey design involving 400 university lecturers, secondary teachers, and adult education facilitators purposively selected from institutions with active digital learning programmes across three Nigerian states. A validated questionnaire (r = 0.76) measured educators’ digital adoption, literacy levels, and perceptions of assessment changes. Data were analysed using descriptive statistics and multiple regression. Results show moderate adoption of digital platforms and widespread but uneven digital literacy, alongside persistent infrastructural disparities. Regression findings reveal that digital platforms (β = .85) and digital literacy (β = .69) positively predict innovative assessment practices, while infrastructure access unexpectedly shows a negative predictive effect (β = –.79), indicating that infrastructure alone does not guarantee effective digital assessment integration. The study concludes that digital competence and purposeful pedagogical alignment are essential for transformative assessment reforms. It recommends strengthened professional development, improved but context-responsive infrastructure, and redesigned assessment frameworks that support continuous, technology-enabled learning.
Keywords: Lifelong learning, Digital literacy, Assessment practices, Digital platforms, Educational technology