📄 social_science

Lecturers’ Use and Perceptions of Machine Learning Informed Educational Interventions for the Early Identification of Learning Difficulties among Students with Intellectual Disabilities in Nigerian Universities

Although lecturers in Nigerian universities perceive machine learning interventions as effective for early identifying learning difficulties and enhancing academic engagement among students with intellectual disabilities, their actual usage remains low due to insufficient training and infrastructure, highlighting a critical gap between perceived value and practical implementation.

Original authors: MOSES ETOMA, TIMOTHY UKAH, ANTHONIA ADIE

Published 2026-07-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: MOSES ETOMA, TIMOTHY UKAH, ANTHONIA ADIE

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a university classroom as a busy, noisy train station. Most passengers (students) know exactly which train to catch and when. But for some passengers with intellectual disabilities, the signs are blurry, the announcements are confusing, and they often miss their trains or get stuck on the wrong platform. These are the "learning difficulties" the paper talks about.

For a long time, the station staff (lecturers) have had to spot these confused passengers by just looking at them and guessing. It's like trying to find a specific person in a crowd by squinting your eyes; it's slow, often wrong, and by the time you realize someone is lost, they've already missed their train.

This research paper asks: What if the station had a smart, digital radar system (Machine Learning) that could spot these confused passengers early, before they even miss their train?

Here is the story of what the researchers found, told in simple terms:

The Big Idea: A "Smart Radar" vs. Reality

The researchers wanted to see if university teachers in Nigeria were using this "smart radar" (Machine Learning tools) to help students with intellectual disabilities. They also asked: Do the teachers think this radar works?

The Reality Check:
The study found that nobody is really using the radar yet.

The Analogy: Imagine the university bought a high-tech, GPS-enabled navigation system for every teacher's desk. But when the researchers looked, they found the screens were turned off. The teachers hadn't been taught how to use it, the internet connection was spotty, and the system wasn't even plugged into the curriculum.
The Result: The teachers' actual use of these tools was "Low." They are still mostly relying on their old-fashioned "squinting" method.

The Surprise: Teachers Love the Idea

Even though the teachers aren't using the tools, they really believe the tools would work if they had them.

The Analogy: It's like a group of drivers who have never driven a self-driving car, but when asked, they all say, "Oh, I'm sure self-driving cars would make the trip smoother, keep us on time, and stop us from getting lost."
The Finding: The teachers rated the potential of these tools very high. They believe that if a computer could analyze student data (like attendance, test scores, and behavior), it could:
- Spot learning trouble spots earlier.
- Keep students more motivated and interested.
- Help students finish their tasks.
- Reduce the stress and anxiety students feel when they are struggling.

The "Who Matters" Test

The researchers wanted to know: Does it matter who the teacher is?

Gender: Does it matter if the teacher is a man or a woman? No. Both groups were equally unlikely to be using the tools.
The University: Does it matter if the teacher works at a rich "Federal" university or a "State" university? No. The type of school didn't change the outcome much.
The Real Hero: The only thing that made a huge difference was whether the teacher actually used the tools.
- The Metaphor: It doesn't matter if you have a Ferrari (a fancy university) or a Toyota (a state university). If you don't press the gas pedal (use the technology), you aren't going to go faster. The study found that the act of using the tool was the strongest predictor of whether students stayed engaged and interested in class.

The Safety Rule: The Radar is a Helper, Not a Doctor

The paper is very careful to explain what this technology is not.

The Analogy: Think of the Machine Learning tool as a smoke detector.
- If the smoke detector beeps, it doesn't mean there is a fire (a clinical diagnosis). It just means, "Hey, check this out, something might be wrong."
- The tool is designed to give a "heads up" so the teacher can say, "I think this student needs a specialist."
- The specialist (a doctor or psychologist) is the one who actually confirms the diagnosis using official rulebooks (like the DSM-5 or ICD-11). The computer never replaces the doctor; it just helps the doctor find the patient faster.

The Bottom Line

The paper concludes that in Nigerian universities, there is a big gap between hope and action.

Hope: Teachers know these smart tools could be amazing for helping students with intellectual disabilities.
Action: They aren't using them because they lack training, the internet is slow, and the schools haven't set up the systems properly.

The Solution: To fix this, the universities need to stop just buying the "smart radar" and start teaching the teachers how to drive it. They need better internet, better training, and a clear plan to connect the computer's "heads up" signals to real help for the students.

In short: The technology is the engine, but the teachers are the drivers. Right now, the engines are sitting in the garage. If we teach the drivers how to start them, the students will finally get the ride they need.

Technical Summary: Lecturers' Use and Perceptions of Machine Learning Informed Educational Interventions for the Early Identification of Learning Difficulties among Students with Intellectual Disabilities in Nigerian Universities

Problem Statement
Despite Nigeria's ratification of international frameworks (UN SDGs 4 and 10) and national policies promoting inclusive education, the implementation of inclusive practices in higher education remains fragmented. Specifically, the early identification of learning difficulties among students with intellectual disabilities relies heavily on manual, episodic, and subjective assessment practices. These conventional methods often result in delayed interventions, marginalization of affected students, and poor educational outcomes. While Artificial Intelligence (AI) and Machine Learning (ML) offer transformative potential for data-driven early screening, their application in Nigerian universities remains largely conceptual. There is a critical gap in understanding how lecturers perceive and utilize ML-informed approaches as decision-support tools to bridge the divide between traditional diagnostic frameworks (DSM-5, ICD-11, AAIDD) and emerging predictive analytics. Furthermore, existing literature lacks empirical evidence on the readiness of lecturers to adopt these tools for early identification and engagement enhancement in this specific context.

Methodology

Research Design: The study employed a descriptive survey research design. Crucially, the authors clarify that ML was utilized strictly as a conceptual and interpretive framework, not as an empirical computational methodology. No ML algorithms were developed, trained, or tested; no datasets were partitioned, and no performance metrics (e.g., accuracy, F1-score) were calculated.
Participants: The population consisted of 600 lecturers from Faculties of Education and Special Needs Departments across 12 selected federal and state universities in the South-South geopolitical zone of Nigeria. Participants were selected via accidental (convenience) sampling, requiring experience with students with intellectual disabilities and familiarity with educational technology.
Instrumentation: Data were collected using the Machine Learning-Based Educational Interventions for Early Identification of Learning Difficulties Questionnaire (MLEIDQ). The instrument included 20 items on a 4-point Likert scale measuring: (i) the extent of ML tool usage for early identification, and (ii) perceptions of ML interventions' impact on early detection and academic engagement.
Data Analysis: Descriptive statistics (mean, standard deviation) and inferential statistics (independent-samples t-tests, multiple linear regression) were analyzed using SPSS version 26. The study tested three null hypotheses regarding gender differences, institutional type differences, and the predictive power of ML use on academic engagement.

Key Results

Extent of Use: Lecturers reported a low level of use (Cluster Mean = 2.37, categorized as "Low Extent") regarding ML-informed digital tools. Specific areas of low adoption included applying ML to individual learning needs (Mean = 1.29), curriculum integration (Mean = 1.58), and routine intervention use (Mean = 1.83). However, lecturers reported higher engagement with institutional support for ML (Mean = 3.29) and tracking student progress (Mean = 3.16).
Perceptions of Effectiveness: Despite low actual usage, lecturers held positive perceptions of ML interventions (Cluster Mean = 3.02, categorized as "Great Extent"). They perceived ML as effective in improving student attention, motivation, communication, and task completion. Approximately 70% of perception indicators exceeded the benchmark mean of 3.0.
Predictive Analysis:
- Hypothesis 1 (Gender): No significant difference was found in ML tool usage based on gender ( $p > 0.05$ ). Gender was not a significant predictor of ML use in the regression model ( $\beta = -0.04, p = 0.398$ ).
- Hypothesis 2 (Institutional Type): No significant difference was found in perceptions of ML support based on university type (Federal vs. State) ( $p > 0.05$ ). Institutional type was not a significant predictor of academic engagement ( $\beta = 0.03, p = 0.450$ ).
- Hypothesis 3 (Prediction of Engagement): The null hypothesis was rejected. Lecturers' use of ML-informed interventions was the strongest and only significant predictor of academic engagement among students with intellectual disabilities ( $\beta = 0.60, p < .001$ ). The regression model explained 38% of the variance in engagement ( $R^2 = 0.38$ ).

Key Contributions

Conceptual Clarification: The study provides a structured conceptual framework illustrating how ML can function as a decision-support mechanism for early screening without replacing formal clinical diagnosis (DSM-5/ICD-11). It delineates the pathway from predictive indicators to professional referral and individualized education planning (IEP).
Empirical Evidence of the "Perception-Use Gap": The research highlights a significant disconnect in Nigerian higher education: while lecturers recognize the potential value of ML for inclusive education, practical implementation is hindered by limited training, infrastructure, and technical competence.
Determinants of Engagement: The study establishes that the practice of using ML tools is a more critical determinant of student academic engagement than demographic factors (gender, rank) or institutional structures (Federal vs. State).

Significance and Claims
The paper claims that its findings contribute to emerging scholarship on AI-enabled inclusive education by providing evidence-based insights into the readiness of lecturers in a developing context. It argues that the adoption of ML is not merely a technological challenge but a capacity-building imperative. The study concludes that for ML to effectively support the early identification of learning difficulties and enhance academic engagement, there must be a strategic focus on:

Strengthening lecturer capacity through AI literacy and data analytics training.
Improving digital infrastructure and institutional support systems.
Developing ethical governance frameworks that integrate ML screening with established diagnostic and intervention pathways.

The authors maintain a modest stance, acknowledging that their findings are based on self-reported data and convenience sampling, limiting generalizability. They emphasize that ML serves as a supplementary tool for screening and referral, not a replacement for professional clinical judgment, and call for future research involving probability sampling and longitudinal empirical evaluations of actual ML systems.