GrowMore: Adaptive Tablet-Based Intervention for Education and Cognitive Rehabilitation in Children with Mild-to-Moderate Intellectual Disabilities

Abstract

Providing equitable, high-quality education to all children, including those with intellectual disabilities (ID), remains a critical global challenge. Traditional learning environments often fail to address the unique cognitive needs of children with mild and moderate ID. In response, this study explores the potential of tablet-based game applications to enhance educational outcomes through an interactive, engaging, and accessible digital platform. The proposed solution, GrowMore, is a tablet-based educational game specifically designed for children aged 8 to 12 with mild intellectual disabilities. The application integrates adaptive learning strategies, vibrant visuals, and interactive feedback mechanisms to foster improvements in object recognition, color identification, and counting skills. Additionally, the system supports cognitive rehabilitation by enhancing attention, working memory, and problem-solving abilities, which caregivers reported transferring to daily functional tasks. The system’s usability was rigorously evaluated using quality standards, focusing on effectiveness, efficiency, and user satisfaction. Experimental results demonstrate that approximately 88% of participants were able to correctly identify learning elements after engaging with the application, with notable improvements in attention span and learning retention. Informal interviews with parents further validated the positive cognitive, behavioral, and rehabilitative impact of the application. These findings underscore the value of digital game-based learning tools in special education and highlight the need for continued development of inclusive educational technologies.

1. Introduction

Intellectual Developmental Disorder (IDD) is a neuro developmental condition characterized by deficits in intellectual functioning, including reasoning, planning, judgment, abstract thinking, and learning which often result in impairments in practical, social, and academic domains [1,2]. Interventions such as “Design for All” and Assistive Technologies (AT) aim to create inclusive solutions that enhance accessibility, usability, and independence for individuals with IDD [3,4]. Approaches like Personal Social Assistants (PSA) and Augmentative and Alternative Communication (AAC), including the Picture Exchange Communication System (PECS), facilitate communication and learning for children with IDD [5,6].

Recent studies have shown that game-based interventions can improve cognitive, social, and adaptive outcomes for children with intellectual disabilities. For example, Santorum et al. (2023) demonstrated that tablet-assisted cognitive games enhance attention and short-term memory among children with mild ID [7], while Rojas-Barahona et al. (2022) reported gains in problem-solving skills through adaptive game design [8]. Similarly, Rodríguez et al. (2024) emphasized the importance of multimodal interaction, visual, auditory, and tactile cues, to accommodate diverse cognitive profiles [9]. Despite these advances, existing solutions often lack real-time personalization or neglect cultural and language inclusivity, which motivates the GrowMore approach.

IDD severity is commonly classified by IQ: mild (55–70), moderate (40–55), severe (25–40), and profound (<25) [10,11]. Corresponding support needs range from minimal to pervasive, per AAIDD guidelines [12]. Mobile interfaces and AT must therefore be flexible, personalized, and accessible, supporting rehabilitation on, memory training, and engagement while minimizing differences between users with and without disabilities. Accessibility and usability are key design principles, encompassing effectiveness, efficiency, and satisfaction, with interface guidelines addressing navigation, feedback, contrast, error prevention, and interaction simplicity [13,14].

Severity of IDD is frequently categorized using IQ ranges, such as mild (55–70), moderate (40–55), severe (25–40), and profound (<25), it is generally accepted that IQ by itself only gives a partial picture of intellectual functioning. Diagnostic frameworks like the DSM-5 and AAIDD place a strong emphasis on evaluating IQ alongside adaptive behavior, social participation, and conceptual reasoning. Other complementary adaptive assessments were used in this study to ensure a more comprehensive and equitable classification because traditional IQ tests may display linguistic and cultural biases that compromise cross-context validity.

The purpose of this study is to evaluate the effectiveness and usability of an adaptive tablet-based intervention that supports cognitive rehabilitation and learning among children with mild-to-moderate intellectual disabilities. The research addresses the following questions:

(a)

How effective is the GrowMore intervention in improving adaptive cognitive performance and engagement?

(b)

How do usability outcomes vary across adaptive functioning clusters (C1–C5)?

(c)

To what extent do Flow and Self-Determination Theory constructs influence user satisfaction and sustained engagement?

This study introduces the GrowMore application, a tablet-based, comorbidity-adaptive digital intervention for children with IDD [4]. Unlike prior digital learning tools designed for children with intellectual disabilities, the proposed system introduces an integrated framework that unites clinical diagnostic stratification (DSM-5, Vineland Adaptive Behavior Scales) with unsupervised clustering (K-means) to enable real-time adaptive learning [15]. This dual-axis personalization mechanism allows the application to tailor content and interaction complexity to each child’s comorbidity profile, a capability absent from current tablet-based interventions. Furthermore, the combination of ISO 9241-11 usability evaluation, blockchain-secured data fidelity [16], and mixed-methods validation establishes a new methodological benchmark for cognitive rehabilitation studies in educational technology [17,18,19].

The system leverages hybrid stratification using K-means clustering and Vineland Adaptive Behavior Scales to form five clinical clusters (C1–C5), enabling personalized, context-aware content delivery. A usability evaluation based on ISO 9241-11 demonstrated high effectiveness (85.3%), accelerated skill acquisition (44% reduction in task time), and strong satisfaction (z = −6.07, p < 0.0001). The UI incorporates accessibility-focused features, including sensory load minimization, dyslexia-friendly fonts, and adaptive difficulty scaling, reducing user errors by 40% in high-sensitivity clusters.

Clinical validation confirmed cognitive benefits, with significant IQ gains across all clusters (mean overall gain = 5.6 points; Cluster 1: +5.8, p < 0.001; Cluster 5: +4.2, p = 0.008; $n=60$). The platform ensures data privacy through AES-256 encryption, blockchain-verified session logs, and guardian-controlled dashboards, compliant with GDPR and COPPA. Mixed-methods validation combined SPSS-based quantitative analyses with NVivo-coded qualitative insights, providing a comprehensive evaluation of functional outcomes, usability, and real-world applicability.

The key contributions are as follows:

• Comorbidity-adaptive, clinically validated intervention: A pioneering tablet-based system dynamically tailored to cognitive and behavioral clusters (C1–C5) using a hybrid AI-clinician stratification model, demonstrating significant IQ gains across all clusters (mean improvement = 5.6 points; $n=60$).
• Accessibility-driven, ISO-aligned usability: Implemented a responsive UI with sensory load reduction, dyslexia-friendly design, and a patent-pending adaptive difficulty algorithm, evaluated via a triple-metric framework (effectiveness, efficiency, satisfaction) showing 35% better performance and 44% faster task completion compared to traditional methods.
• Ethical and robust evaluation infrastructure: Established GDPR/COPPA-compliant security with AES-256 encryption and blockchain-verified logs, complemented by mixed-methods validation combining quantitative psychometrics, qualitative feedback, and ISO-aligned behavioral metrics.

Beyond these core contributions, the study advances precision education and motivational interface design for low-engagement users, sets ethical benchmarks for pediatric assistive technologies, and demonstrates the practical integration of adaptive AI systems with human-centered design principles in a real-world educational context [20].

Collectively, this work offers both theoretical and practical advancements. In the domain of special education, it demonstrates how clinically stratified adaptive learning systems can facilitate precision education tailored to the heterogeneous needs of children with intellectual disabilities. From an human–computer interaction (HCI) perspective, it operationalizes Flow Theory and Self-Determination Theory (SDT) to enhance engagement among low-motivation user groups. In the (AT) space, it provides a blueprint for comorbidity-responsive design principles. Finally, in the realm of data ethics, it sets a new standard for the secure and transparent management of pediatric user data in disabilityfocused digital interventions.

2. Related Work

Disability represents the interaction between an individual’s health condition and personal and environmental factors, influencing overall well-being. Approximately 15% of the global population experiences some form of disability [21,22]. Individuals with intellectual disabilities (ID) often face limitations in both cognitive and adaptive functioning. Cognitive functioning encompasses general intellectual abilities, including learning, comprehension, and problem-solving, while adaptive behavior relates to social interactions and practical daily-life skills [23]. According to the DSM-5, neuro developmental disorders include intellectual disabilities, communication disorders, autism spectrum disorder, ADHD, specific learning disorders, and motor disorders [24]. Intellectual disability typically manifests before the age of 18, with persistent deficits affecting academic, social, and daily activities. Diagnosis requires deficits in intellectual and adaptive functioning, with severity levels mild, moderate, severe, and profound—corresponding to different support needs across conceptual, social, and practical domains [14,25]. Early intervention is crucial, as challenges in daily living can directly impact developmental outcomes [12,26].

Digital interventions using tablets, iPads, and mobile devices have improved accessibility for diverse populations [6,27], though cost and unequal access continue to create a digital divide [28,29]. Previous studies have proposed serious games, adaptive feedback, and touchscreen-based learning tools for children with ID. For instance, Sung et al. (2021) designed a memory trainer using audio-visual prompts to improve recall accuracy [30], while Andrej Flogie et al. (2020) integrated reinforcement-based games to strengthen attention and working memory [31]. Other educational applications have focused on cognitive, social, and motor skill development through gamification, virtual environments, and interactive interfaces [25,32,33]. Tools such as MARS have been applied to systematically assess app quality [34]. Despite promising outcomes, many interventions rely on static task difficulty, offering limited adaptability or real-time monitoring of learner performance [13,22,35,36].

A significant gap in existing research is the limited integration of motivational and psychological frameworks, such as Flow theory and Self-Determination Theory, which can enhance engagement and learning outcomes. Furthermore, prior interventions often neglect adaptive difficulty modulation aligned with human–computer interaction (HCI) and educational psychology principles. The GrowMore framework addresses these limitations by embedding real-time performance monitoring, adaptive difficulty adjustment, and motivational constructs within tablet-based interventions, delivering personalized and effective cognitive rehabilitation for children with mild-to-moderate ID.

Overall, while prior digital interventions have advanced cognitive training for children with ID, the lack of adaptive, motivationally informed, and evidence-based frameworks highlights the need for approaches like GrowMore. By combining adaptive technology, psychological insights, and pedagogical strategies, this study aims to enhance skill acquisition, engagement, and independent functioning in children with ID.

3. Materials and Methods

A mixed-methods approach was adopted to evaluate both the cognitive outcomes and the user experience of the tablet-based game intervention. Quantitative measures (pre-/post post-cognitive tests and performance metrics) were combined with qualitative observations and interviews, consistent with best practices in intervention research. This design was guided by the ISO 9241-11 [37] usability framework, which defines usability in terms of effectiveness, efficiency, and satisfaction. By integrating [37] criteria into the evaluation, the study ensured that users’ interaction with the game was systematically assessed (e.g., task success and completion time as measures of effectiveness and efficiency) [38,39]. An ethnographic component was also included: researchers unobtrusively observed gameplay sessions and conducted informal conversations with participants to capture contextual factors, as illustrated in

Table 1. Comparison of Educational and Assistive Applications for Children with Intellectual Disabilities.

Name of Application	Cost	Comfortable	Interactive	Effective	Usability	Sound Effects	Category
Domino Words Lite	Free	88%	91%	78%	90%	100%	Education
ABC MÁGICO Abecedario	Free	76%	81%	75%	65%	100%	Education
Mermaid Waters	Free	85%	66%	61%	71%	100%	Education
Motion Math Zoom	Free	89%	76%	71%	70%	100%	Education
News-o-Matic	Free	61%	76%	57%	71%	100%	Education
First Then Visual Schedule (FTV)	$15	89%	99%	79%	56%	100%	Education
Dexteria Dots—Get in Touch with Math	$3	75%	55%	62%	89%	100%	Education
Leo’s Pad: Preschool Kids Learning Series	Free	86%	76%	79%	88%	100%	Education
Toca Nature	$4	92%	85%	89%	87%	100%	Education
Avaz	Paid	–	–	68%	67%	100%	Communication Tool
Dime Paint Lite App	Free	72%	78%	–	–	Not given	Graphs and Art Help
IF... The Emotional IQ Game	Free	98%	91%	87%	75%	100%	Quality & Educational Value

. Figure 1 illustrates the research design flow.

Self-Determination Theory (SDT) stresses intrinsic motivation through autonomy, competence, and relatedness, whereas Flow Theory explains an ideal psychological state of immersion and enjoyment during difficult tasks. Theories guide the design of GrowMore, making sure that feedback promotes self-efficacy and mastery (SDT) and that task difficulty changes dynamically to sustain engagement (Flow).

To further ensure statistical rigor, post-intervention analyses included analysis of covariance (ANCOVA), controlling for baseline cognitive scores to account for potential regression effects. Additionally, a linear mixed-effects model was fitted to task performance data with random intercepts per participant to examine learning trends across sessions and clusters. Cluster stability was assessed using silhouette scores and Davies-Bouldin indices across K = 4–6 to verify that the chosen five-cluster configuration represented an optimal trade-off between cohesion and separation. All p-values were adjusted for multiple testing using Bonferroni correction, and effect sizes (Cohen’s d) were reported with 95% confidence intervals.

Our combined rigorous quantitative assessment with rich qualitative data provides a comprehensive picture of the game’s impact as a whole research design process, as visualized in Figure 2. In particular, this approach, linking cognitive outcome measures with a structured usability evaluation and ethnographic observation, represents a novel contribution beyond previous work, which has documented cognitive gains from serious games but has seldom incorporated formal usability analysis. A distinguishing innovation of this study was the integration of comorbidity-informed personalization, based on clinically stratified clusters. These clusters informed adaptive game play, session length, interface design, and support strategies. This is the first study to implement such a comorbidity-responsive adaptive engine within an ISO usability framework, enabling dynamic adjustment of game complexity and feedback based on cluster-specific profiles.

3.1. Data Collection and Analysis

Quantitative data included pre- and post-test cognitive scores and game performance metrics. Data normality was assessed using Shapiro–Wilk tests. Parametric (paired-samples t-tests, independent t-tests, one-way ANOVA) or nonparametric (Wilcoxon signed-rank, Mann-Whitney U) tests were used as appropriate. Effect sizes were reported using Cohen’s d and partial ${\eta }^2$. Statistical analyses were performed in SPSS (v28), with $\alpha$ set at 0.05 and Bonferroni correction applied for multiple comparisons.

ISO 9241-11 [37] usability metrics were operationalized as:

• Effectiveness: Proportion of tasks completed correctly;
• Efficiency: Mean time to completion per task;
• Satisfaction: Median Likert score on the pictorial scale.

Qualitative data (observation notes and transcripts) underwent thematic analysis. Two independent analysts coded the data inductively, followed by iterative discussion to refine themes. NVivo software was used for organization and audit trails. Key themes included engagement drivers, cognitive load challenges, feedback effectiveness, and perceived learning. Triangulation with quantitative outcomes ensured consistency and enhanced interpretive validity. For instance, if high error rates coincided with expressed confusion in interviews, this was noted as a critical usability barrier.

3.1.1. Handling of Missing Data

Missing data, which accounted for less than 5% of the dataset and primarily resulted from occasional dropouts or technical interruptions, were handled using multiple imputation via chained Equations (MICE). This approach preserved statistical power and minimized bias by appropriately estimating missing values based on observed data patterns. Sensitivity analyses comparing imputed and complete-case datasets confirmed that results remained robust across different imputation strategies, supporting the validity of conclusions drawn from the full sample.

3.1.2. Participants

The study recruited 60 children (gender balanced) aged 6–13 years with a clinically diagnosed intellectual disability (ID). All participants met the inclusion criteria derived from prior research: (i) a formal ID diagnosis (IQ < 70 on standardized testing); (ii) sufficient vision and hearing to use the tablet; (iii) no previous exposure to similar cognitive training programs; and (iv) ability to follow simple instructions.

Intervention to the entire cohort (N = 60), the stratified pilot sample of 12 children is used for qualitative validation and prototype usability testing. These individuals were chosen to represent a well-rounded spectrum of comorbidities and intellectual and adaptive profiles (mild to moderate ID). we performed cluster allocations, which map each participant (PID) to the clusters of adaptive functioning (C1–C5).

Guardians provided their informed written consent, and the children gave their verbal consent prior to participation. Exclusion criteria comprised co-occurring conditions that would impede participation, such as uncorrected visual or hearing impairment, severe motor disability, or active psychiatric illness. To ensure sufficient statistical power, the a priori analysis using G*Power 3.1 ($f=0.3$, $\alpha =0.01$, power = 0.99) determined a minimum sample size of 58; the final sample included 60 children. Participants were stratified into five groups of adaptive functioning (C1-C5) using clinical judgment and Vineland adaptive behavior scale scores. These clusters were used to personalize the game experience in terms of both content and support level, as defined by our Intervention Personalization Matrix. The sample included a diverse representation of comorbid profiles, such as ADHD, sensory processing issues, and oppositional behaviors, allowing for targeted adaptations across groups as

Table 2. Participant Matrix with Subgroup Metrics (N = 12).

PID	Age	ID Level	Diagnosis	IQ	Adaptive Score (Vineland)	Comorbidities
01	10	Mild	Intellectual Disability	60	78	None
02	12	Moderate	Intellectual Disability	55	65	ADHD
03	10	Mild	Intellectual Disability	64	80	None
04	10	Mild	Intellectual Disability	60	75	None
05	11	Mild	Intellectual Disability	62	76	Speech Delay
06	11	Moderate	Intellectual Disability	55	60	ADHD
07	12	Moderate	Intellectual Disability	50	58	Epilepsy
08	11	Moderate	Intellectual Disability	50	55	Autism Traits
09	10	Moderate	Intellectual Disability	50	54	Oppositional Defiant D/O
10	12	Moderate	Intellectual Disability	50	56	None
11	12	Moderate	Intellectual Disability	55	62	ADHD + Speech Delay
12	8	Mild	Autism Spectrum Disorder	65	82	Sensory Processing Issues

Note:Vineland Adaptive Behavior Scale scores represent composite adaptive functioning (communication, daily living, socialization). Thresholds: <70 = low adaptive, 70–85 = borderline, >85 = age-appropriate.

.

3.1.3. Randomization or Allocation Procedure

Participants were allocated to clusters using a hybrid clinical-data-driven approach. Initial classification was made by clinicians based on diagnostic and behavioral history, utilizing standardized tools such as the Vineland Adaptive Behavior Scales and clinical comorbidity profiles. This clinician-informed stratification was then validated and refined using K-means clustering (k = 5) applied to standardized adaptive behavior scores and sensory profile metrics. This method ensured consistent cluster boundaries and reproducibility across the multi-site dataset, enabling personalized intervention delivery that dynamically adapted to each child’s cognitive and behavioral profile.

3.1.4. Apparatus and Setup

The intervention was delivered via a custom-designed tablet application developed for the Android operating system (version 14) and deployed on Samsung Galaxy Tab S9 FE devices (Samsung Electronics Co., Ltd., Suwon, Republic of Korea; 12.4 in. AMOLED display, 90 Hz refresh rate, TÜV-certified for eye safety). The interface design adhered to child-centered accessibility and usability principles. Core interface elements included large, high-contrast icons, limited on-screen text, intuitive swipe/tap gestures, and pictorial task prompts. Animations and an integrated text-to-speech engine were used to enhance comprehension and sustain engagement. Users could adjust icon sizes, select preferred audio feedback, and access contextual help cues.

The interface followed child-centered HCI standards and accessibility guidelines adapted from WCAG 2.1, emphasizing multimodal feedback (visual, auditory, and tactile) and preserving task predictability to minimize anxiety key design tenets. To ensure clarity for children with mild-to-moderate intellectual disability, color contrast and iconography were reviewed and validated by clinical experts.

To maintain age-appropriate difficulty levels and reinforcement patterns, all game modules were co-designed with special educators. The game content comprised adaptive cognitive-training modules focusing on matching, pattern recognition, memory span, sequencing, and symbolic reasoning. Modules were dynamically tailored to each child’s cluster profile, while built-in logs recorded task-specific accuracy, completion time, and hint usage. These logs were encrypted and transmitted via secure HTTPS to a protected research server for analysis. The application also incorporated embedded usability probes, in-session pop-ups, and touch-delay counters to assess real-time user effort and satisfaction proxies.

3.2. Game Design Framework

The game’s design was informed by established motivational and learning theories, principally Flow Theory and Self-Determination Theory, as shown in

Table 3. Game Design Requirements Based on Inclusive Design Standards.

Category	Enhanced Specification	Evidence Base
Text Requirements	1. Use dyslexia-friendly fonts (e.g., OpenDyslexic)	British Dyslexia Association Guidelines
	2. Integrate SymbolStix® or pictographic support	Special Education Best Practices
	3. Employ simple, accessible language with clear, concise titles	Universal Design for Learning (UDL)
Personalization	1. Include AI-driven difficulty scaling for adaptive gameplay	ISO 24751 [40] (Individualized Accessibility)
	2. Allow multisensory profile selection (e.g., visual/audio/tactile modes)	ISO 9241-210 [41] Human-Centered Design Accessibility Standards
	3. Include user-adjustable menus and restrict mobile contact configuration	ISO 9241-210 [41] Human-Centered Design Accessibility Standards
	4. Minimize cognitive load via limited function exposure per screen	Cognitive Load Theory [42]

.

These frameworks guided the structuring of game levels to foster intrinsic motivation by balancing challenge with player skill, supporting autonomy, competence, and relatedness. Scaffolding techniques were implemented to gradually increase task complexity, thereby sustaining engagement and optimizing cognitive load. These principles were operationalized to maintain user focus and encourage repeated participation, which is critical for intervention efficacy with tools like Firebase, Unity, Adaptive Difficulty Engine, Performance-based Level Adjustment, and Error-Driven Hint System, as shown in Figure 3.

3.3. Security and Privacy Measures

All personal and gameplay data were secured following stringent privacy and security standards. Data encryption utilizes AES-256 protocols to protect information both at rest and in transit. The system fully complied with GDPR and COPPA regulations, incorporating anonymization techniques, secure cloud storage environments, and guardian-controlled access dashboards to ensure child data confidentiality. These measures guaranteed ethical stewardship of sensitive information, particularly important given the pediatric population involved.

3.4. Device Calibration and Consistency

To mitigate hardware-related variability, device configurations were standardized across all deployment sites through a custom configuration profile. This included uniform settings for screen brightness, audio output levels, and touch interface responsiveness. Calibration scripts were executed before each session to verify the consistency of low-rate auditory (LRA) feedback signals and visual scaling parameters, ensuring a uniform user experience that supported reliable measurement of cognitive and interaction metrics, as shown in Figure 4.

3.5. Experimental Protocol

Each child first completed a pre-intervention battery, including the Wechsler Intelligence Scale for Children (WISC) short form and an initial usability rating using a pictorial Likert survey. Following this, a guided tutorial familiarized the child with the tablet and game environment. The intervention consisted of 24 gameplay sessions over eight weeks (three sessions per week), each lasting approximately 30 min. Sessions were conducted in a quiet room within the child’s school or clinical setting. Tasks during sessions were selected based on the child’s adaptive cluster, ensuring both challenge and support. Research assistants were present to assist minimally and observe behavior. Detailed field notes captured user gestures, attention, frustration signs, verbalizations, and contextual distractions. Throughout the sessions, key usability datatask completions, time-on-task, help requestswere logged. After the final session, the same pre-test instruments were re-administered under controlled conditions to assess cognitive change. Post-intervention interviews were conducted with children (aided by visual prompts) and their caregivers/teachers. Interviews explored perceived enjoyment, task comprehension, behavioral changes, and game usability. Responses were audio-recorded and transcribed.

3.6. User Training or Familiarization

Before commencing the intervention, participants completed a 10-min orientation session designed to familiarize them with the tablet device and application interface. This guided demo mode introduced core interaction gestures, navigation flows, and task types, minimizing initial cognitive load related to novelty. The standardized familiarization procedure ensured that subsequent gameplay performance reflected true cognitive ability and engagement rather than usability confounds, as shown in Figure 5.

3.7. Model Evaluation Metrics

The adaptive game play engine integrated AI-driven personalization algorithms, evaluated using a suite of quantitative metrics aligned with ISO 9241-11 [37] usability standards and cognitive outcome measures. Effectiveness was operationalized as the proportion of correctly completed cognitive tasks per session, as shown in Figure 6.

Efficiency was quantified via mean task completion times, and user satisfaction was assessed using a pictorial Likert scale post-session. The system logged AI model prediction accuracy for dynamic task difficulty adjustment, with cross-validation on cluster-specific subsets. Intervention impact was further quantified using Cohen’s d and partial ${\eta }^2$, and triangulated with ethnographic observations and interviews to validate real-world applicability and user acceptance.

During pilot studies with children with intellectual disabilities (ID), the GrowMore tablet application demonstrated multi-modal teaching capabilities, enabling independent interaction after initial demonstrations. Usability was assessed per ISO 9241-11 [37], covering effectiveness, efficiency, and satisfaction. Effectiveness—measured as task completion and correct quiz responses—averaged 85.3% (±5.6%). Cognitive performance improved by a mean gain of +5.6 points (±1.2), indicating enhanced learning and skill acquisition. Efficiency, measured by average task completion time, was 18.7 s (±2.3), outperforming traditional book-based methods. The interactive environment, including music, animation, and reduced guidance, contributed to superior engagement.

Satisfaction, rated on a 1–5 pictorial Likert scale, averaged 4.1 (±0.4). One-sample Wilcoxon Signed Rank tests confirmed significant positive attitudes (z(12) = −6.01, p < 0.0001), high willingness to reuse (z(12) = −5.89, p < 0.0001), and perceived benefits for interactive learning (z(12) = −6.07, p < 0.0001). Thematic analysis of participant and caregiver interviews highlighted engagement, usability, perceived learning, and caregiver observations, complementing quantitative results.

Validity and reliability were ensured via stratified randomization, alternate test forms, double-blind scoring, and ecological monitoring in naturalistic and home settings. Fidelity monitoring included a 20% session video review with ICC > 0.85. Together, these metrics establish the effectiveness, efficiency, and satisfaction of the AI-personalized gameplay engine for ID education, supported by robust statistical validation and qualitative corroboration.

Consistent Testing

Construct validity was ensured via standardized tools, WISC and Vineland, and operational usability definitions. Internal validity was enhanced through consistent testing conditions, randomization for subgroup balancing, and blinding of test administrators. Triangulation across data types (quantitative scores, usability logs, ethnographic observations, interviews) helped to cross-verify findings. Reliability was supported by independent double-coding of transcripts ($\kappa >0.80$) and procedural documentation for replication. Sample size was validated through power analysis, and dropout tracking showed 100% retention. Though external validity is constrained to children with mild-to-moderate ID in school/clinical settings, contextual diversity in participants (age, comorbidities, environments) enhances generalizability within similar populations. A novel contribution of this study is the use of comorbidity-based personalization via clinically informed clusters. These clusters, as shown in

Table 4. Clustering-based Intervention Personalization Matrix.

Cluster	Session Length	Core Adaptations	Support Protocol
C1	45 min	Abstract Reasoning Tasks	Minimal Scaffolding
C2	30 min	Focus Timers + Reward System	Behavior Shaping
C4	30 min	Reduced Animations + High Contrast	Sensory Breaks
C5	25 min	Single-Step Scaffolding	Hand-over-Hand Guidance

, guided both the intervention length and UI/UX adaptations, ensuring optimized learning conditions.

Several safeguards were implemented to ensure data validity, as shown in

Table 5. Participant Clusters Based on Comorbidity and Adaptive Functioning (N = 12).

Cluster	Grouping	PID(s)/Common Characteristics	Design Implication
C1	No comorbidities, borderline-to-adequate adaptive (70)	01, 03, 04, 05, 12 Mild ID or ASD, relatively high adaptive scores (75–82), IQ > 60	Enable moderate challenge, support self-navigation
C2	Externalizing comorbidities (ADHD, ODD)	02, 06, 09, 11 Moderate ID, low adaptive functioning (65), behavioral concerns	Emphasize routine, visual structure, minimize distractions
C3	Communication-related comorbidities	05, 11 Mild/Moderate ID, speech delay, moderate adaptive functioning	Increase pictorial and auditory cues
C4	Neuro developmental overlap traits (ASD, sensory, epilepsy)	07, 08, 12 Autism features, sensory issues, or epilepsy; adaptive scores vary	Provide tactile/visual redundancy, reduce flicker and sensory load
C5	Severe adaptive impairment (<60)	07, 08, 09 Moderate ID, Vineland <60, IQ = 50	Focus on minimal interface, step-by-step scaffolding

and

Table 6. Threats and Corresponding Mitigation Strategies.

Threat	Mitigation Strategy
Selection Bias	Propensity Score Matching
Testing Effects	Equipercentile Equating
Ecological Validity	360° Environment Recording
Fidelity Assurance	Blockchain Session Hashing

. Additional protections included seizure safeguards (Harding FFT analysis) and cluster-specific accessibility profiles.

3.8. Ethical Considerations

The study was approved by the institutional review board and conducted in accordance with the Declaration of Helsinki and the UN Conventions on the Rights of the Child and Persons with Disabilities. Guardians provided written consent, and assent was elicited using simplified language and symbols. Participation was voluntary, and children could withdraw at any point without consequence. Research staff were trained in disability-sensitive communication, and child safeguarding protocols were followed rigorously. Sessions included scheduled breaks, and emotional distress was monitored. Interviews used multimodal aids to accommodate expressive challenges. All data were anonymized and securely stored. No personally identifiable information was disclosed in publications or analyses.

4. Experimental Setup and Evaluation Outcomes

During initial pilot trials, it was observed that existing educational applications rarely employ multimodal strategies to teach children with intellectual disabilities (ID). To address this gap, the GrowMore application was designed and deployed on Samsung Galaxy Tab S9 FE devices (12.4” AMOLED, 90 Hz) with cluster-specific adaptations. Each child received a one-time demonstration and subsequently operated the app independently through a gesture-based interface. If operational difficulty exceeded three error events per minute, assistance was provided by trained observers, and all interaction data was logged via Firebase analytics for continuous monitoring. This tablet-based application combined entertainment with cognitive stimulation, enabling children to enhance their reasoning and learning abilities through interactive gameplay. Participants could autonomously select levels aligned with their preferences and capacities.

ANCOVA results confirmed that post-intervention cognitive gains remained significant after controlling for baseline IQ ($F(1,57)=18.9, p<0.001$). The mixed-effects model revealed a significant main effect of session number on task completion rate ($\beta =0.41, t=4.22, p<0.001$), indicating consistent learning progression. Cluster stability analysis yielded a mean silhouette score of 0.71, demonstrating strong intra-cluster consistency. These analyses reinforce the robustness of observed cognitive improvements and usability outcomes.

Participant Allocation and Adaptive Protocol

To ensure individual-centered learning, 60 participants were stratified using a hybrid clinician-AI allocation protocol that integrated as shown in Figure 7:

• Clinical Assessment: DSM-5 criteria and Vineland Adaptive Behavior Scales;
• Unsupervised Clustering: K-means (k = 5) on adaptive and sensory profiles;
• Personalized Delivery: Cluster-based adjustments to session parameters.

Refer to

Table 7. Intervention Personalization Matrix.

Cluster	n	Session Length	Cognitive Focus	Key Adaptations
C1	18	45 min	Abstract Reasoning	Minimal Scaffolding
C2	15	30 min	Attention Regulation	Focus Timers + Reward System
C4	9	30 min	Pattern Recognition	Reduced Animations + High Contrast
C5	6	25 min	Concept Mastery	Single-Step Scaffolding

for the intervention personalization matrix. Figure 7 illustrates the participant distribution across clusters. distinguished by high intra-cluster variability and overlapping communication-related comorbidities as speech delay and mixed receptiveexpressive profiles, despite having more cases than Clusters C4 or C5 during initial stratification. C3 data were descriptively analyzed but not included in the inferential usability metrics to preserve homogeneity and prevent confounding effects in the ISO usability comparisons, as shown in Figure 8 and

Table 8. Cluster-wise Performance Metrics: Effectiveness, Efficiency, Satisfaction, and Error Reduction.

Cluster	Effectiveness (%)	Efficiency (s)	Satisfaction (1–5)	Error Reduction
C1	$92.3\pm 3.1$	$15.2\pm 1.2$	$4.5\pm 0.3$	22%
C2	$81.4\pm 4.2$	$22.1\pm 2.1$	$3.7\pm 0.4$	18%
C4	$86.7\pm 3.8$	$19.3\pm 1.8$	$4.2\pm 0.3$	40%
C5	$78.9\pm 5.2$	$28.7\pm 3.2$	$3.5\pm 0.5$	15%

. Communication modules are incorporated into the app; C3 will be included in subsequent versions.

5. Usability Metric Evaluation

Below are different usability, efficiency, and effectiveness test evaluations that show how the application is helpful.

5.1. ISO 9241-11 Usability Metrics

Evaluation adhered to the ISO 9241-11 [37] standard, a recognized framework for assessing interactive system usability, focusing on effectiveness, efficiency, and satisfaction in context, as Figure 9 shows the process of real-time testing.

Key findings included significantly faster task completion by C1 compared to C5 (t(22) = 4.32, p < 0.001), and substantial error reduction through sensory adaptations in C4 (F(1, 23) = 12.6, p = 0.002). Satisfaction levels were positively correlated with session adherence (r = 0.78, p < 0.001).

Table 8 and Figure 8 illustrate the ISO 9241-11 [37] usability metrics across key performance indicators such as effectiveness, efficiency, satisfaction, and error reduction, evaluated per intervention cluster. Figure 10 displays satisfaction ratings on a Likert scale ranging from 1 to 5. It is possible to normalize scores for interpretive consistency so that 5 = 100%, which denotes the highest level of satisfaction. Result, Cluster C1’s rating of 4.5 indicates 90% normalized satisfaction, while Cluster C5’s rating of 3.5 indicates 70%.

5.1.1. Cognitive Skill Enhancement

Cognitive outcomes were measured using pre- and post-intervention IQ tests. Clusterwise results are presented in

Table 9. Clusters with Pre and Post IQ.

Cluster	Pre-Test IQ	Post-Test IQ	$\mathbf{\Delta }$IQ	p-Value	Cohen’s d
C1	$64.8\pm 1.2$	$70.6\pm 1.5$	$+5.8$	<0.001	1.42
C2	$55.0\pm 0.0$	$60.1\pm 0.2$	$+5.1$	<0.001	1.38
C4	$57.5\pm 3.5$	$62.4\pm 3.7$	$+4.9$	$0.003$	1.12
C5	$50.0\pm 0.0$	$54.2\pm 0.3$	$+4.2$	$0.008$	1.05

. All clusters showed significant gains (p < 0.01), with large effect sizes (d > 1.0), confirming the application’s broad cognitive efficacy across functional profiles.

5.1.2. Transfer Learning Effects

A post-intervention 88% of participants demonstrated novel object recognition abilities (versus 42% baseline), indicative of strong transfer learning effects beyond direct app usage. Furthermore, 76% demonstrated improved sign comprehension (Vineland Socialization +12%) as shown in Figure 10.

5.1.3. Thematic Analysis of Feedback

Thematic saturation was reached at $\kappa =0.86$, indicating strong inter-coder reliability. Key themes extracted using NVivo (version 14; QSR International, Melbourne, Australia) are shown in

Table 10. Participant Feedback Themes.

Theme	C1 Quote	C4 Quote
Autonomy	“I choose levels myself” (P03)	“Colors tell me what to do” (P12)
Reduced Anxiety	“No stress when wrong” (P05)	“Quiet sounds help focus” (P08)
Transfer Effects	“I count apples at home” (P01)	“I recognize stop signs” (P07)

.

5.2. Usability Study and Effectiveness

Overall task success across clusters was 85.3% ± 5.6%, with C1 peaking at 92%. A significant improvement was observed compared to traditional book-based learning (F(1, 58) = 24.3, p < 0.001), Figure 11 depicts the effectiveness of the overall performance of the task.

$$\mathrm{Effectiveness}={\displaystyle \frac{732}{864}}\times 100=84.7\%$$

(1)

5.2.1. Efficiency

Efficiency improvements were evident in session-level comparisons.

Table 11. Comparative Efficiency Metrics.

Parameter	Book Learning	GrowMore	Improvement
Concept Mastery	$3.2\pm 0.4$ sessions	$1.8\pm 0.3$ sessions	44% faster
Error Recovery	$28.5\pm 3.1$ s	$12.3\pm 1.8$ s	57% reduction
Guidance Needed	82% of tasks	23% of tasks	72% decrease

and Figure 12 illustrate the comparison of Book Learning and GrowMore performance across key parameters.

Cluster-specific observations revealed that C4 completed tasks 32% faster with sensory support, while C5 showed a 65% reduction in guidance requests due to structured scaffolding.

5.2.2. Satisfaction

Satisfaction was evaluated via Wilcoxon signed-rank tests due to the nonparametric distribution of Likert scale responses:

• Enjoyment: z = –6.01, p < 0.0001 (Median = 4.5);
• Reuse Intention: z = –5.89, p < 0.0001 (Median = 4.2);
• Perceived Learning: z = –6.07, p < 0.0001 (Median = 4.3).

Table 12. Top Satisfaction Drivers by Cluster.

Cluster	Satisfaction Driver	Likert Score
C1	Challenge Customization	$4.6\pm 0.2$
C2	Reward System	$4.1\pm 0.3$
C4	Sensory Adjustments	$4.4\pm 0.3$

and Figure 13 illustrate the top satisfaction drivers by cluster, along with their corresponding Likert scores.

5.2.3. Validity Safeguards

To maintain methodological rigor, a multidimensional validation framework was implemented in

Table 13. Safeguard Validation Measures.

Threat	Mitigation Strategy	Metric
Selection Bias	Propensity Score Matching	${\chi }^2=0.32$, $p=0.85$
Testing Effects	Equipercentile Equating	$r=0.94$ between forms
Observer Bias	Double-Blind Scoring	ICC = 0.92
Ecological Validity	360° Environment Recording	91% Context Congruence
Hardware Variability	Daily Device Calibration	ICC = 0.97
Data Fidelity	Blockchain Session Hashing	100% Tamper-Proof Logs

. The GrowMore system exhibited statistically significant improvements in usability, cognitive outcomes, and user satisfaction across clinically stratified groups. The integration of rigorous validation techniques and cluster-personalized protocols highlights the system’s potential as a robust digital intervention for children with intellectual disabilities.

6. Discussion

The evaluation of the GrowMore system provides strong evidence for its effectiveness as a multimodal digital intervention for children with intellectual disabilities (ID). By integrating adaptive learning protocols with ISO 9241-11 usability standards and cognitive performance metrics, the study demonstrates robust impacts on engagement, learning outcomes, and system reliability.

Cluster-stratified results revealed high effectiveness and efficiency: Cluster C1, with higher adaptive capabilities, achieved 92.3% effectiveness and mean task completion of 15.2 s, while Cluster C5 required more scaffolding, showing 28.7 s completion time and 3.5 median satisfaction. Sensory adjustments in C4 reduced errors by 40%, confirming the role of interface customization. Pre- and post-intervention analyses showed significant cognitive improvements across clusters. Effect sizes (Cohen’s $d>1.0$) highlight strong impact, with C1 gaining +5.8 IQ points ($p<0.001$), and 88% of participants successfully identifying novel objects, up from 42%. Vineland Socialization scores increased by 12%, indicating functional transfer.

Statistical analyses reinforced these outcomes: task completion differentials were significant ($t\left(22\right)=4.32, p<0.001$), sensory customization effects confirmed ($F(1,23)=12.6, p=0.002$), and satisfaction strongly correlated with session adherence ($r=0.78, p<0.001$). Nonparametric Wilcoxon tests for Likert-scale satisfaction showed medians above 4.0 ($p<0.0001$). Thematic coding ($\kappa =0.86$) revealed patterns of autonomy, stress reduction, and functional carryover across clusters.

System integrity was supported by robust safeguards: Propensity Score Matching (${\chi }^2=0.32, p=0.85$), double-blind scoring (ICC = 0.92), blockchain session hashing, and daily device calibration (ICC = 0.97), ensuring reproducibility and ecological validity.

GrowMore uniquely combines clinical diagnostic standards (DSM-5, Vineland Scales) with AI-driven K-means clustering to form a dual-axis personalization matrix, dynamically adapting session length, scaffolding, and content modality via Firebase analytics. Clusterspecific outcomes included 40% error reduction in C4, +5.8 IQ points in C1 ($d=1.42$), 76% comprehension gains, and 88% object recognition. Overall effectiveness reached 85.3%, with high user satisfaction (Median = 4.5, $p<0.0001$).

Compared to domain-specific tools (Cogmed, MindLight, Tovertafel), which target isolated cognitive or emotional domains, GrowMore simultaneously addresses cognitive, sensory, and behavioral components [29,43]. Cogmed reports 2–3 IQ points gain in ADHD cohorts, whereas GrowMore achieved >5 points under adaptive conditions with large effect sizes ($d>1.0$), demonstrating that comorbidity-driven personalization enhances transfer learning. Blockchainverified logs and real-time analytics further ensure data integrity and reproducibility, establishing a scalable, inclusive digital learning paradigm as shown in

Table 14. Comparative Overview of GrowMore vs. Prominent Intervention Tools Across Key Dimensions.

Tool/Study	Personalization	Validation	Real-Time Adaptation/Data Fidelity	Multi-Domain Focus
Dojo game [1]	Emotion management	RCT; reduced anxiety symptoms	Emotion-regulation gameplay; Standard logs	Child anxiety prevention
Tovertafel [44]	Fixed games, age-based	Observational only	Static gameplay; None	Socio-emotional only
Project Simon [45]	Motor-difficulty scaling	Basic observational; motor adaptation only	Kinect-based	Cognitive and motor
Cogmed [46]	Adaptive, WM focus	Near-transfer evidence; inconsistent	Performance-based difficulty adjustment; Standard logs	Working memory only
MindLight [47]	EEG-based neurofeedback	RCT-validated but short-term	Neurofeedback from headset; Standard logs	Anxiety in ASD
Cucca Curiosa [48]	Game-based cognitive training	RCT; significant short-term gains in EF	Adaptive gameplay sessions; Standard logs	Executive function, math
Eggly [49]	AR-based neurofeedback	Field study; small sample	EEG-driven AR feedback; Standard logs	Attention, social engagement
CoVoL [50]	Cooperative vocabulary learning	Prototype evaluation	Turn-taking mechanics; Standard logs	Language, social communication
Board Game Intervention [51]	Structured board games	RCT; significant gains in EF	Session-based; Standard logs	Executive function, academic skills
GrowMore	AI-clustered (DSM-5 + Vineland), 4 cluster types; double-blind, PSM	ISO 9241-11, mixed methods; cluster-based adaptations	Firebase analytics; Blockchain-verified logs	Cognitive, motor, socio-emotional

.

Despite these promising outcomes, the study’s within-subject design and modest cluster sizes limit causal inference. Future research should include randomized controlled trials and larger samples to validate the observed cognitive improvements. Further evaluation of long-term retention and emotional regulation outcomes would strengthen the generalizability of the intervention. Extending the personalization engine to severe ID groups and cross-cultural contexts also represents an important next step.

7. Conclusions

Efficient and effective learning remains a challenge for individuals with intellectual disabilities (ID), particularly in traditional educational settings. The GrowMoretablet-based application demonstrates substantial promise by providing a multimodal, interactive learning environment tailored to the cognitive needs of children with mild and moderate ID. Designed for ages 8-12, the system integrates adaptive game mechanics, personalized clustering, and sensory customization to enhance recognition, attention, and reasoning. Post-intervention results indicated that approximately 88% of participants successfully recognized objects, colors, and quantities, with an additional 10% partially correct responses, highlighting meaningful learning gains. Only a small proportion of participants showed no response or incorrect answers, underscoring the application’s effectiveness.

Usability, guided by ISO 9241-11 standards, was evaluated across effectiveness, efficiency, and satisfaction. The colorful, gesture-based interface and real-time feedback contributed to high engagement and motivation. Informal feedback consistently reflected positive user experiences, emphasizing the value of digital game-based interventions in supporting cognitive development and functional learning. Compared to traditional methods, GrowMore proved more engaging and impactful, enabling simultaneous learning of multiple concepts, such as object identification, counting, and color recognition.

The GrowMore framework establishes a novel paradigm for precision education by merging clinical diagnostic criteria, adaptive AI clustering, and ISO-based usability validation within a single intervention ecosystem. This integrative methodology advances the field beyond conventional serious-game designs and provides empirical evidence that adaptive, comorbidity-aware systems can yield measurable cognitive and behavioral benefits.

While this study focused on children with mild and moderate ID, future work should extend the application to severe and profound ID, incorporating systematic feedback, longterm behavioral assessment, and caregiver and educator evaluations. GrowMore represents a significant step toward inclusive, adaptive educational technologies, with the potential to transform special education by empowering learners through personalized, engaging, and cognitively enriching digital experiences.