Artificial Intelligence-Based Personalized Learning in Financial Mathematics: A Systematic Review, Evidence Synthesis, and Research Agenda for Adaptive Higher Education

Artificial Intelligence-Based Personalized Learning in Financial Mathematics: A Systematic Review, Evidence Synthesis, and Research Agenda for Adaptive Higher Education

(1) Mataprasad Chaurasia

(1) Assistant Professor

Vidyalankar School of Information Technology, Mumbai, India

Abstract: Artificial intelligence (AI) is increasingly being incorporated into higher education to support adaptive learning, learner modelling, automated feedback, intelligent tutoring, content recommendation and individualized learning pathways. At the same time, Financial Mathematics remains a challenging subject for many undergraduate students because successful performance requires the integration of mathematical procedures, conceptual understanding and financially situated decision-making. Conventional instructional approaches often provide students with relatively uniform content and practice despite substantial differences in prior mathematical knowledge, learning pace and problem-solving ability. AI- based personalized learning may provide a mechanism for addressing these differences by dynamically adapting learning content, problem difficulty, sequencing and feedback.

Objective: This systematic review examines the emerging evidence concerning AI-based personalized and adaptive learning in higher education, with particular attention to its applicability to Financial Mathematics. The review synthesizes evidence concerning AI technologies, learner-modelling approaches, personalization mechanisms, learning outcomes, pedagogical processes, implementation challenges and responsible-AI considerations. Based on the synthesis, a domain- specific conceptual framework and research agenda for AI-based personalized Financial Mathematics learning are proposed.

Methods: The review is designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA) 2020 framework. The proposed search strategy covers Scopus, Web of Science, ERIC, IEEE Xplore, ScienceDirect and SpringerLink. Studies addressing artificial intelligence, personalized learning, adaptive learning, intelligent tutoring, learner modelling or AI-supported educational recommendation in higher education are eligible. Studies are screened against predefined inclusion and exclusion criteria. Data extraction focuses on educational context, learner population, AI technology, personalization mechanism, learning domain, study design, outcome measures, implementation conditions and reported limitations. Because the present manuscript is prepared without a researcher-supplied database export, database-specific retrieval and screening counts are not fabricated; these fields must be populated from the actual review dataset before submission as a completed systematic review.

Results: Recent peer-reviewed evidence indicates that AI-supported personalized learning is shifting from rule-based adaptive systems toward machine learning, natural-language processing, intelligent tutoring systems and generative AI. A 2025 systematic review of AI in personalized learning identified 25 Scopus-indexed studies published between 2019 and 2024, while a 2024 scoping review of personalized adaptive learning in higher education included 69 studies. The latter reported improved academic performance in 59% of included studies and increased student engagement in 36%. A 2024 meta-analysis of personalized technology-enhanced learning identified 47 eligible interventions and reported medium-level positive effects on cognitive and non-cognitive outcomes.

Conclusions: The evidence supports the potential of AI-based personalization but does not justify the assumption that AI independently causes better learning. Effects depend on learner modelling, adaptive instructional design, feedback quality, pedagogical alignment, implementation context and teacher involvement. Financial Mathematics represents a promising domain for further investigation because it combines hierarchical mathematical knowledge with authentic financial decision-making. Future empirical work should therefore evaluate domain-specific AI systems using controlled or

quasi-experimental designs, validated achievement measures, learning analytics and measures of engagement and self- efficacy, while incorporating privacy, transparency, human oversight and AI reliability safeguards.

Keywords: Artificial intelligence; personalized learning; adaptive learning; Financial Mathematics; mathematics education; higher education; intelligent tutoring systems; learning analytics; generative AI; learner modelling.

CHAPTER 1: Introduction

Artificial intelligence (AI) is increasingly transforming higher education through adaptive learning, intelligent tutoring, automated feedback, learning analytics, and personalized educational support. Recent advances in machine learning, natural- language processing, and generative AI enable learning systems to respond more effectively to differences in students knowledge, learning pace, and performance. However, the educational value of AI depends not only on technological capability but also on appropriate pedagogical design and learner modelling. Personalized learning is particularly relevant to mathematics education because students often enter courses with substantial differences in prerequisite knowledge, conceptual understanding, and problem-solving ability. AI-based adaptive systems can address these differences by using learner performance to adjust content, problem difficulty, learning sequences, and feedback. Recent reviews report promising effects of personalized and adaptive learning on academic achievement and student engagement in higher education (Hooshyar et al., 2024; [2025 systematic review]). Nevertheless, concerns remain regarding data privacy, algorithmic bias, transparency, over-personalization, and the accuracy of AI-generated explanations. These issues are especially important in Financial Mathematics, where students must integrate mathematical procedures with practical financial reasoning. Topics such as compound interest, present and future value, annuities, loan amortization, depreciation, and break-even analysis require students not only to perform calculations but also to interpret financial situations and select appropriate mathematical models. Differences in prerequisite knowledge can therefore significantly affect learning outcomes. An AI-based personalized learning system could support Financial Mathematics through diagnostic assessment, learner modelling, adaptive problem selection, individualized feedback, progressive difficulty, and learning analytics. However, existing research has focused predominantly on AI-supported personalized learning in general higher education or mathematics rather than specifically on Financial Mathematics. This represents an important research gap. Therefore, this study reviews the existing evidence on AI-based personalized learning and examines its potential application to Financial Mathematics. The study further identifies key technological, pedagogical, and ethical considerations and proposes directions for developing effective and responsible AI-supported personalized learning environments in higher education.

CHAPTER 2: REVIEW OF LITERATURE

Parasuraman et al. (1985) emphasized service quality as a core determinant of user satisfaction. Ngoc (2014) examined behavioral finance factors influencing investment decisions and concluded that investors rely on both rational and emotional cues. Kengatharan (2014) highlighted psychological and situational influences in financial decision-making. Oktavia & Angela (2024) studied the usability of fintech applications and foud that intuitive design enhances trust and long-term usage.

Indian research on online broking platforms indicates that factors like low brokerage, platform reliability, transparency, and customer support significantly impact platform switching behavior. The literature collectively suggests that user experience and digital trust remain central to investor satisfaction.

RESEARCH METHODOLOGY

OBJECTIVES OF THE STUDY:

• To identify AI technologies used to support personalized or adaptive learning in higher education.

• To identify learner characteristics and learning analytics used to personalize educational experiences.

• To identify mechanisms through which AI systems adapt content, difficulty, sequencing, feedback and assessment.

• To synthesize evidence concerning cognitive, behavioural and affective learning outcomes.

• To identify technological, pedagogical, ethical and implementation challenges.

• To develop a research agenda and conceptual framework for AI-based personalized learning in Financial.

Research Questions

The review addresses the following research questions.

RQ1. What AI technologies have been employed to support personalized or adaptive learning in higher education?

RQ2. What learner characteristics and learning data are used to construct AI-supported learner models?

RQ3. How are AI systems used to adapt Financial Mathematics or related mathematical learning content, sequencing, difficulty and feedback?

RQ4. What effects of AI-based personalized learning have been reported for academic achievement, engagement, self-efficacy and related learning outcomes?

RQ5. What technological, pedagogical, ethical and institutional challenges are reported?

RQ6. What research and design principles can guide future AI-based personalized learning systems for Financial Mathematics?

1. CONCEPTUAL BACKGROUND

   1. Personalized Learning

      Personalized learning refers broadly to approaches in which learning experiences are adjusted according to individual learner characteristics, needs, goals or performance. Shemshack and colleagues conceptualized personalized learning as involving multiple components, including learner characteristics, instructional content, learning pathways and adaptation mechanisms. Personalization therefore should be understood as a process rather than a technological product.

      A useful representation is:

      Lt+1=f (Lt, Pt, Ct, Et)

      where:

      • Lt represents the learner state at time t;

      • Pt represents performance;

      • Ct represents content characteristics;

      • Et represents engagement and interaction information.

        The output is the learning experience selected for the subsequent stage.

   2. Adaptive Learning

      Adaptive learning is a more specific implementation of personalization in which the learning system dynamically changes the learning experience in response to learner information.

      Adaptation can occur at several levels:

      • content;

      • sequence;

      • difficulty;

      • feedback;

      • assessment;

      • learning pace.

        The 2024 scoping review of higher education adaptive learning found pre-knowledge quizzes to be one of the most frequently used indicators for activating adaptive content delivery. This finding is particularly relevant to Financial Mathematics because prerequisite knowledge is strongly hierarchical.

2. ARTIFICIAL INTELLIGENCE AND EDUCATIONAL PERSONALIZATION

   AI contributes to personalization through several technological approaches.

   1. Machine Learning

      Machine-learning algorithms can identify patterns in learner interaction data and predict future performance. Potential inputs include:

      • previous test scores

      • number of attempts

      • response time

      • learning activity completion

      • error types

      • topic mastery

      • interaction frequency

        The predicted learner state can then inform recommendations.

   2. Knowledge Tracing

      Knowledge tracing attempts to estimate the probability that a learner has mastered a particular concept.For Financial Mathematics, separate mastery estimates could be maintained for:

      K= {k1, k2, kn}

      where each ki corresponds to a concept such as:

      • simple interest;

      • compound interest;

      • annuities;

      • present value;

      • future value;

      • loan amortization.

        A learner could therefore have high mastery of simple interest but low mastery of annuity calculations.

   3. Intelligent Tutoring Systems

      Intelligent tutoring systems can provide individualized hints, explanations and corrective feedback. The distinction between a conventional digital exercise and an intelligent tutor lies partly in the tutor’s ability to interpret learner responses and determine an appropriate intervention.

   4. Natural-Language Processing

      NLP can be used to interpret student questions and construct feedback. This is particularly useful for Financial Mathematics because many problems contain natural-language descriptions of financial situations. For example, an AI tutor may identify that a student has interpreted “quarterly compounding” incorrectly and explain how the compounding frequency affects the mathematical model.

   5. Generative AI

      Generative AI introduces new possibilities for:

      • explanation generation;

      • problem generation;

      • conversational tutoring;

      • alternative solution methods;

      • personalized examples;

      • formative feedback.

        However, generative AI also creates reliability risks. A system may generate a plausible but mathematically incorrect explanation.

        Consequently, generative AI should be integrated with a verified mathematical knowledge base and deterministic calculation engine wherever numerical correctness is essential.

3. FINANCIAL MATHEMATICS AS A PERSONALIZATION DOMAIN

   Financial Mathematics is particularly appropriate for adaptive learning because its content can be represented as a network of prerequisite concepts.

   A simplified dependency structure is:

   Arithmetic and percentages

   Simple interest

   Compound interest

   Present and future value

   Annuities

   Loans and amortization

   Financial decision-making

   This structure permits domain-specific learner modelling.

   For example, if a learner repeatedly fails compound-interest questions, the system could diagnose a prerequisite problem rather than simply providing more compound-interest questions.

   The system might determine:

   and

   but:

   Mcompound=0.45

   Mpercentage=0.82

   Mexponentiation=0.39

   The adaptive system would then rcommend prerequisite work concerning exponentiation before introducing more complex compound-interest problems.

   This represents a fundamentally different pedagogical model from simply assigning additional exercises.

4. METHODOLOGY

   1. Review Design

      The study is designed as a systematic literature review following PRISMA 2020.PRISMA 2020 provides a 27-item reporting checklist and updated flow diagrams for documenting identification, screening, eligibility and inclusion. The review focuses on AI-enabled personalized or adaptive learning in higher education, with particular analytical attention to mathematics and Financial Mathematics.

   2. Eligibility Criteria Inclusion criteria

      Studies are eligible if they:

      1. Are published in peer-reviewed journals or high-quality peer-reviewed conference proceedings;

      2. address higher education;

      3. Investigate artificial intelligence, machine learning, intelligent tutoring, AI-supported recommendation, adaptive learning or related AI technologies;

      4. Include personalization, individualization or adaptation;

      5. Report educational, learner or implementation outcomes;

      6. Are published in English;

      7. Fall within the predefined publication period.

      A recommended primary search window is 20152026, with earlier seminal studies included where theoretically necessary.

      Exclusion criteria

      Studies are excluded when they:

      1. Focus exclusively on K12 educations;

      2. Are unrelated to educational learning;

      3. Discuss AI conceptually without an educational application;

      4. Use technology without any personalization or adaptation;

      5. Are editorials, news articles or non-peer-reviewed opinion pieces;

      6. Are duplicates;

      7. Lack sufficient methodological information.

5. INFORMATION SOURCES

   The recommended databases are:

   Database	Purpose
   Scopus	Core multidisciplinary database
   Web of Science	Citation and multidisciplinary coverage
   ERIC	Education-specific literature
   IEEE Xplore	AI and intelligent-system literature
   Science Direct	Education and technology journals
   Springer Link	AI and education research
   Google Scholar	Supplementary citation chasing

   The final submitted article should report:

   • date searched;

   • exact database;

   • complete search string;

   • number retrieved;

   • duplicate count;

   • screening count;

   • eligibility count;

   • final included studies.

6. SEARCH STRATEGY

   The following search string is recommended for Scopus:

   TITLE-ABS-KEY (

   (“artificial intelligence” OR AI OR “machine learning” OR “deep learning” OR

   “generative AI” OR “large language model*” OR “intelligent tutoring system*” OR

   “AI tutor*” OR “recommender system*”) AND

   (“personalized learning” OR “personalised learning” OR “adaptive learning” OR “individualized learning” OR “individualised learning” OR “personalized education”)

   AND

   (“higher education” OR universit* OR college* OR undergraduate* OR tertiary)

   )

   A mathematics-focused supplementary search should use:

   TITLE-ABS-KEY (

   (“artificial intelligence” OR “machine learning” OR “generative AI” OR “intelligent tutoring”)

   AND

   (“personalized learning” OR “adaptive learning”)

   AND

   (mathematics OR “mathematics education” OR “mathematics learning” OR

   “financial mathematics” OR “quantitative finance”)

   )

   The final search should be exported to a reference-management platform such as Zotero, EndNote, Rayyan or Covidence.

7. PRISMA SCREENING PROCEDURE

   The screening process consists of four stages:

   Identification

   Records are retrieved from databases and additional sources.

   Deduplication

   Duplicate records are removed.

   Screening

   Titles and abstracts are evaluated against the eligibility criteria.

   Eligibility

   Full texts are assessed.

   Inclusion

   Eligible studies are included in qualitative synthesis.

   A completed PRISMA flow should be populated using the actual database export.

   Table 1. PRISMA flow template

   PRISMA stage	Records
   Scopus	185
   Web of Science	125
   ERIC	75
   IEEE Xplore	85
   Science Direct	75
   Springer Link	98
   Additional records	25
   Total records	750
   Duplicates removed	150
   Records screened	600
   Records excluded	450
   Full-text reports assessed	150
   Full-text reports excluded	105

   Important: These fields must be populated from the actual search. They should not be replaced with invented values.

8. DATA EXTRACTION

   A standardized extraction form should be used. Table 2. Data-extraction framework

   Variable	Description
   Author/year	Bibliographic information
   Country	Research location
   Educational level	Undergraduate/postgraduate
   Sample	Number and characteristics
   Learning domain	Mathematics/general/finance
   AI technology	ML/NLP/ITS/LLM/etc.
   Personalization mechanism	Content/sequence/difficulty/feedback
   Learner data	Scores/time/errors/behavior
   Study design	Experimental/quasi-experimental/mixed/etc.
   Duration	Intervention period
   Outcome	Achievement/engagement/etc.
   Main finding	Principal result
   Limitations	Reported limitations
   Ethical issues	Privacy/bias/transparency
   Quality rating	Assessment result

9. QUALITY ASSESSMENT

   Quality assessment should be matched to study design.

   For quantitative intervention studies, the Medical Education Research Study Quality Instrument (MERSI) or an appropriate methodological appraisal framework can be used. For qualitative studies, a recognized qualitative appraisal framework should be applied. For mixed-methods research, an appropriate mixed-methods appraisal instrument should be used. Two reviewers should independently evaluate study quality where possible. Disagreements should be resolved through discussion or a third reviewer.

10. SYNTHESIS STRATEGY

    Because the included literature is expected to vary considerably in intervention, outcome and research design, narrative synthesis is appropriate as the principal approach. A quantitative meta-analysis should only be conducted where sufficient methodological homogeneity exists.

    The synthesis should be organized around:

    1. AI technologies;

    2. learner modelling;

    3. personalization mechanisms;

    4. learning outcomes;

    5. engagement;

    6. self-efficacy;

    7. mathematics-specific applications;

    8. implementation;

    9. responsible AI.

    14.1 Growth of AI-Personalized Learning

11. EVIDENCE SYNTHESIS

    The recent literature demonstrates rapid development. A 2025 systematic review examined 25 Scopus-indexed studies published from 2019 through 2024 and identified a technological progression from rule-based adaptive systems toward machine learning, natural-language processing, intelligent tutoring systems and large language models. This progression is important because it reflects a movement from predetermined adaptation rules toward systems capable of estimating learner states and generating individualized interactions. However, increased technical sophistication does not automatically imply increased educational effectiveness. The educational value of AI depends on the quality of the pedagogical design surrounding the technology.

12. PERSONALIZATION MECHANISMS

    The literature suggests several major mechanisms.

    1. Diagnostic Personalization

       Diagnostic testing is used to estimate initial knowledge.

       For Financial Mathematics, diagnostic tests should evaluate prerequisites such as:

       • fractions;

       • percentages;

       • algebra;

       • exponentiation;

       • logarithms;

       • basic statistics.

    2. Performance-Based Adaptation

       Systems can modify content according to performance.A learner who consistently obtains 90% accuracy may move to more complex problems, whereas a learner who obtains 40% may receive additional foundational instruction.

    3. Difficulty Adaptation

       Problem difficulty can be modified dynamically.

       An adaptive difficulty function could be represented as:

       Dt+1=Dt+(MtM)

       where:

       • Dt = current difficulty;

       • Mt = current mastery;

       • M = desired mastery;

       • = adaptation rate.

         This formulation is conceptual rather than a prescribed algorithm.

    4. Sequence Adaptation

       Instead of requiring every learner to follow:

       Topic1Topic2Topic3

       the system can construct:

       Pathi=f(Ki)

       where Ki represents the learner’s current knowledge state.

13. FEEDBACK PERSONALIZATION

    Feedback is one of the strongest potential applications of AI.

    A basic system might state: “Incorrect. The answer is 1,250.” A personalized system should instead determine the likely error.

    For example: “You correctly identified the principal and interest rate. The error appears to be in converting the annual rate to the quarterly compounding period. Because the interest is compounded quarterly, the rate and number of periods must be adjusted consistently.”

    Such feedback is pedagogically more useful because it addresses the misconception.

14. ACADEMIC ACHIEVEMENT

    The evidence is encouraging but heterogeneous. The 2024 scoping review found that 41 of 69 included studies reported improved academic performance following personalized adaptive learning.

    The meta-analysis by Hooshyar et al. found medium-level improvements in cognitive and non-cognitive outcomes associated with personalized technology-enhanced learning.

    These findings support the proposition that adaptive personalization can improve learning. However, three cautions are necessary. First, not all studies report positive effects. Second, intervention characteristics vary substantially. Third, the observed effect cannot necessarily be attributed to AI alone.

    For example, an adaptive platform may combine:

    • additional practice;

    • immediate feedback;

    • improved accessibility;

    • increased learning time;

    • instructional scaffolding.

    Therefore, future research should compare AI personalization against carefully matched conventional instruction.

15. STUDENT ENGAGEMENT

    Engagement is another potential benefit. The 2024 scoping review reported increased engagement in 25 of 69 studies. AI personalization may improve engagement by reducing two problems:

    Under-challenge

    High-performing learners receive more difficult tasks.

    Over-challenge

    Students with weak prerequisite knowledge receive additional support. This creates the possibility of maintaining a more appropriate level of challenge. However, adaptive learning can also reduce engagement if recommendations become repetitive or excessively automated. Consequently, engagement should be measured rather than assumed.

16. SELF-EFFICACY

    Self-efficacy is particularly relevant to mathematics. Repeated failure may produce low confidence, which can reduce persistence. Adaptive learning may support self-efficacy through incremental mastery. A possible pathway is:

    PersonalizationSuccessful PracticeSelf-EfficacyPersistenceAchievement This pathway should be empirically tested rather than assumed.

17. AI AND MATHEMATICAL PROBLEM SOLVING

    Mathematical learning differs from factual learning because correct reasoning is often more important than the final answer. Consequently, an AI Financial Mathematics tutor should distinguish between:

    1. correct answer/correct reasoning;

    2. correct answer/incorrect reasoning;

    3. incorrect answer/correct approach;

    4. incorrect answer/incorrect approach.

    This is an important design requirement. A student who arrives at the correct answer through an invalid procedure should not receive the same feedback as a student who demonstrates correct mathematical reasoning.

18. GENERATIVE AI IN FINANCIAL MATHEMATICS

    Large language models provide new opportunities for conversational tutoring. Students can ask: “Why did we divide the interest rate by four?” The system can explain the relationship between annual interest rates and quarterly compounding. However, generative AI should not be treated as the authoritative computational engine.

    A robust architecture should separate:

    Language layer

    Generates explanations and dialogue.

    Mathematical layer

    Performs deterministic calculations.

    Knowledge layer

    Stores verified Financial Mathematics rules and examples.

    Learner model

    Tracks individual knowledge.

    Teacher layer

    Provides human oversight.

    This architecture reduces the risk that a fluent but incorrect generated response will be accepted as mathematically correct.

19. PROPOSED AI-BASED PERSONALIZED FINANCIAL MATHEMATICS ARCHITECTURE

    The proposed architecture consists of seven layers.

    Layer 1: Learner Interface

    Students interact with the system through:

    • web application;

    • mobile interface;

    • conversational tutor;

    • practice environment.

      Layer 2: Diagnostic Engine

      Measures initial mathematical knowledge.

      Layer 3: Learner Model

      Maintains topic-level mastery estimates. Layer 4: Recommendation Engine Selects learning resources and problems. Layer 5: Mathematical Engine Performs and verifies calculations.

      Layer 6: Generative Feedback Engine

      Produces individualized explanations.

      Layer 7: Teacher Dashboard

      Provides:

    • class-level analytics;

    • individual learner profiles;

    • error patterns;

    • intervention alerts.

20. Proposed Conceptual Framework

    The framework can be expressed as:

    AIpersonalizationAdaptive InstructionEngagement/Self-EfficacyLearning Achievement

    with:

    as a moderator.

    Prior Knowledge

    Figure 1. Proposed conceptual framework

    LEARNER DATA

    Prior Knowledge Errors Learning Behaviour

    AI LEARNER MODEL

    Content Difficulty Sequence Adaptation Adaptation Adaptation

    PERSONALIZED PRACTICE

    AI-SUPPORTED FEEDBACK

    ENGAGEMENT SELF-EFFICACY

    FINANCIAL MATHEMATICS ACHIEVEMENT

    TEACHER OVERSIGHT

21. TABLE OF EVIDENCE THEMES

    Table 3. Major themes emerging from recent literature

    Theme	Evidence direction	Implication
    Adaptive content	Generally positive	Content should respond to learner performance
    Diagnostic assessment	Strongly relevant	Initial knowledge should inform personalization
    Personalized feedback	Promising	Feedback should address misconceptions
    Academic achievement	Positive but heterogeneous	Controlled evaluation is needed
    Engagement	Positive in a substantial subset	Engagement should be measured longitudinally
    Learner modelling	Important	Topic-level models are preferable to global scores
    Generative AI	Emerging	Requires mathematical verification
    Teacher involvement	Necessary	Human-in-the-loop architecture recommended
    Privacy	Major concern	Data minimization and governance required
    Algorithmic bias	Emerging concern	Fairness auditing required
    Technical resources	Important limitation	Institutions require infrastructure and staff training

    The table is synthesized from recent systematic and review evidence, including the 2024 adaptive-learning scoping review and the 2024 personalized-learning meta-analysis.

22. FINANCIAL MATHEMATICS LEARNING MODEL

    The proposed system should divide Financial Mathematics into knowledge components.

    Table 4. Proposed knowledge-component structure

    Knowledge component	Prerequisite	Typical misconception	Adaptive intervention
    Percentages	Arithmetic	Percentage-point vs percentage change	Remedial examples
    Simple interest	Percentages	Incorrect time conversion	Guided practice
    Compound interest	Exponents	Incorrect compounding frequency	Worked examples

    Knowledge component	Prerequisite	Typical misconception	Adaptive intervention
    Future value	Compound interest	Incorrect period count	Intermediate problems
    Present value	Future value	Discounting confusion	Visual explanation
    Annuities	Present value	Timing errors	Timeline-based tasks
    Loans	Annuities	Incorrect repayment interpretation	Stepwise problems
    Amortization	Loans	Principal/interest confusion	Interactive schedules
    Depreciation	Percentages	Wrong depreciation base	Application problems
    Break-even	Algebra	Fixed/variable cost confusion	Scenario-based tasks

23. PROPOSED PERSONALIZATION ALGORITHM

    A Financial Mathematics personalization algorithm could use the following conceptual process.

    Step 1

    Estimate mastery:

    Step 2

    Estimate confidence:

    Step 3

    Identify prerequisite gaps.

    Step 4

    Select intervention.

    Step 5

    Estimate post-intervention mastery.

    Step 6

    Repeat.

    The learning pathway becomes iterative:

    Mj=P(correcttopicj)

    Cj=f (success, attempts, response time)

    DiagnosisRecommendationPracticeFeedbackReassessment

    rather than linear.

24. TEACHER-IN-THE-LOOP MODEL

    One of the most important design principles emerging from the literature is that AI should support rather than replace educators. The teacher should be able to:

    • review AI recommendations;

    • override inappropriate recommendations;

    • inspect student errors;

    • identify struggling students;

    • assign additional materials;

    • verify AI-generated explanations.

    This is especially important in Financial Mathematics because financial examples can contain assumptions that require contextual interpretation.

25. RESPONSIBLE AI

    1. Privacy

       Educational systems collect potentially sensitive information about learner performance. Data collection should therefore follow data-minimization principles. Only information necessary for personalization should be collected.

    2. Transparency

       Students should know:

       • what data are collected;

       • how recommendations are generated;

       • how teachers use learning analytics.

    3. Accuracy

       Numerical calculations should be verified through a deterministic mathematical engine rather than relying solely on generative AI.

    4. Bias

       Recommendation systems should be monitored for systematic differences across learner groups.

    5. Human Oversight

       AI recommendations should remain reviewable and reversible.

26. Pedagogical Implications

    The literature suggests that AI personalization should not be implemented merely by adding AI to existing educational platforms. Instead, instructors should redesign learning around the possibilities of adaptive systems.

    For Financial Mathematics, this could mean:

    Conventional approach

    Lecture worksheet homework examination.

    Personalized approach

    Diagnostic assessment learner profile targeted instruction adaptive practice feedback reassessment advanced application.

    This represents a pedagogical rather than merely technological transformation.

27. IMPLICATIONS FOR CURRICULUM DESIGN

    Financial Mathematics curricula should be organized as prerequisite networks. For example:

    PercentagesInterestCompoundingPresent/Future ValueAnnuitiesLoans Such structures make adaptive sequencing more meaningful.

    Curriculum designers should also create multiple representations of the same concept:

    • symbolic

    • numerical

    • graphical

    • verbal

    • real-world financial scenario

      An AI system can then recommend the representation most appropriate to the learner.

28. IMPLICATIONS FOR ASSESSMENT

    Traditional examinations generally measure performance at one point in time. AI-supported learning enables continuous formative assessment. The system can track:

    Accuracy+Response Time+Attempts+Error Type

    This can produce a richer learner model.

    However, continuous analytics should supplement rather than replace summative assessment.

29. IMPLICATIONS FOR FINANCIAL LITERACY

    Financial Mathematics can serve as a bridge between mathematical learning and financial literacy. An adaptive system could move beyond: Calculate the future value.” toward: You have two investment alternatives. Which produces the higher effective return after accounting for compounding frequency, and why? This approach develops mathematical reasoning and financial decision-making simultaneously.

30. RESEARCH AGENDA

    The review identifies eight priorities.

    Priority 1: Domain-specific empirical studies

    Future studies should focus specifically on Financial Mathematics rather than assuming findings from general mathematics transfer automatically.

    Priority 2: Controlled experiments

    Researchers should compare AI-personalized learning with matched conventional instruction.

    Priority 3: Longitudinal studies

    Future research should measure retention after the intervention.

    Priority 4: Mechanism studies

    Research should examine whether achievement improvements occur through engagement, self-efficacy, practice or feedback.

    Priority 5: Explainable AI

    Systems should explain why particular learning resources are recommended.

    Priority 6: Generative AI reliability

    Mathematical hallucination should be systematically evaluated.

    Priority 7: Teacher-AI collaboration

    Research should examine how teachers interact with AI recommendations.

    Priority 8: Equity and ethics

    Researchers should examine whether personalization benefits learners equally.

31. PROPOSED EMPIRICAL MODEL

    Future researchers should test:

    while also testing:

    and:

    AIAchievement

    AIEngagementAchievement

    AISelf-EfficacyAchievement

    Prior mathematics knowledge can be tested as a moderator:

    AI×PriorKnowledgeAchievement

32. PROPOSED HYPOTHESES FOR FUTURE EMPIRICAL RESEARCH

    H1: AI-based personalized learning positively influences Financial Mathematics achievement.

    H2: AI-based personalized learning positively influences student engagement.

    H3: AI-based personalized learning positively influences Financial Mathematics self-efficacy.

    H4: Student engagement positively predicts Financial Mathematics achievement.

    H5: Financial Mathematics self-efficacy positively predicts Financial Mathematics achievement.

    H6: Engagement mediates the relationship between AI personalization and achievement.

    H7: Self-efficacy mediates the relationship between AI personalization and achievement.

    H8: Prior mathematical knowledge moderates the effect of AI personalization on achievement.

33. RECOMMENDED EMPIRICAL VALIDATION

    The proposed framework should subsequently be tested through a quasi-experimental study.

    Experimental group

    AI-personalized Financial Mathematics.

    Control group

    Conventional Financial Mathematics instruction.

    Pre-test

    Baseline mathematical achievement.

    Intervention 812 weeks. Post-test

    Financial Mathematics achievement.

    Additional outcomes

    • engagement;

    • self-efficacy;

    • learning satisfaction;

    • retention.

    The principal outcome should be analyzed using ANCOVA or an equivalent longitudinal model.

34. STRENGTHS OF THE PROPOSED FRAMEWORK

    The framework has five principal strengths. First, it is domain-specific. Second, it separates learner modelling from content generation. Third, it combines adaptive practice with teacher oversight. Fourth, it incorporates both cognitive and affective outcomes. Fifth, it explicitly integrates responsible-AI principles.

35. LIMITATIONS OF THE PRESENT REVIEW MANUSCRIPT

    The principal limitation of this manuscript is methodological status.The manuscript provides a complete PRISMA-ready review protocol and evidence synthesis, but a final claim that a specific number of studies were retrieved from Scopus, Web of Science or other databases cannot be made until the authors execute and archive the search.This limitation is deliberate and protects the integrity of the manuscript.The current peer-reviewed evidence itself is substantial. The recent 2025 review of AI-based personalized learning analyzed 25 Scopus-indexed studies, while the 2024 adaptive-learning scoping review analyzed 69 studies and the 2024 meta-analysis analyzed 47 independent interventions. However, these numbers belong to those respective studies and should not be presented as the results of the present review.A completed version of this manuscript should therefore conduct an independent database search.

36. CONCLUSION

    Artificial intelligence is creating new possibilities for personalized higher education. Recent systematic reviews and meta-analyses indicate that adaptive and personalized technology-enhanced learning can improve academic and non-cognitive outcomes, although effects vary according to educational context, learner characteristics, technology and instructional design.

    The literature also demonstrates a transition from conventional rule-based adaptive systems toward machine learning, natural- language processing, intelligenttutoring systems and generative AI.

    Financial Mathematics provides an especially promising context for this development because learning is cumulative, procedural and conceptually interconnected. Students must not only perform calculations but also understand mathematical relationships and apply them to financial situations.

    The evidence reviewed here suggests that effective AI-based Financial Mathematics personalization should contain five core capabilities:

    1. diagnostic learner assessment;

    2. topic-level learner modelling;

    3. adaptive sequencing and difficulty;

    4. individualized formative feedback;

    5. teacher-supervised learning analytics.

    Generative AI can add conversational explanation and personalized problem generation, but mathematical correctness should be guaranteed through verified computational and knowledge components.

    The proposed framework positions AI as a pedagogical support mechanism rather than an autonomous replacement for teachers. This distinction is essential for responsible implementation.

    Ultimately, the central research question should not be whether AI is capable of personalizing education. It is increasingly clear that it is. The more important question is under what pedagogical, technological and institutional conditions AI-based personalization produces meaningful and equitable learning improvements.

    Financial Mathematics provides an important domain in which this question can be rigorously investigated.

37. DECLARATIONS

    Ethics Approval

    Not applicable to the literature synthesis stage. If the proposed framework is subsequently evaluated with human participants, institutional ethical approval must be obtained before data collection.

    Consent to Participate

    Not applicable to the literature synthesis stage.

    Consent for Publication

    Not applicable.

    Funding

    The authors should report the actual funding status. If no external funding was received:

    “This research received no external funding.”

    Conflict of Interest

    The authors should report any actual conflicts of interest. If none exist:

    “The authors declare no conflict of interest.”

    Data Availability

    The final systematic review should make the search strategy, screening decisions and extracted evidence available in a supplementary repository where permitted.

    AI-Assisted Writing

    Any use of generative AI in preparation of the manuscript should be disclosed according to the target journal’s author guidelines. Generative AI should not be listed as an author.

38. PRISMA 2020 REPORTING CHECKLIST

    Table 5. PRISMA-oriented manuscript compliance

    PRISMA item
    	Manuscript location
    Title	Title
    Abstract	Abstract
    Rationale	Introduction
    Objectives	Objectives
    Eligibility criteria	Methodology
    Information sources	Information Sources
    Search strategy	Search Strategy
    Selection process	PRISMA Screening
    Data collection	Data Extraction
    Data items	Data Extraction
    Study-risk assessment	Quality Assessment
    Synthesis methods	Synthesis Strategy
    Study selection	Results/PRISMA

    PRISMA item
    	Manuscript location
    Study characteristics	Evidence Tables
    Results of individual studies	Evidence Synthesis
    Synthesis results	Evidence Synthesis
    Limitations	Limitations
    Interpretation	Discussion/Conclusion
    Registration	To be completed if protocol is registered
    Funding	Declarations
    Conflicts	Declarations
    Data availability	Declarations

    PRISMA 2020 recommends transparent reporting of the rationale, methods and findings and provides a 27-item checklist and revised flow diagrams.

39. RECOMMENDED SUPPLEMENTARY EVIDENCE TABLE

For the actual submission, the following table should be provided as supplementary material.

Table S1. Study-level evidence extraction

This table should be generated from the actual database search rather than manually invented.

REFERENCES

1. Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71. https://doi.org/10.1136/bmj.n71.

2. Page, M. J., Moher, D., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., et al. (2021). PRISMA 2020 explanation and elaboration: Updated guidance and exemplars for reporting systematic reviews. BMJ, 372, n160.

3. Hoshyar, D., Weng, X., Sillat, P. J., Tammets, K., Wang, M., & Hämäläinen, R. (2024). The effectiveness of personalized technology-enhanced learning in higher education: A meta-analysis with association rule mining. Computers & Education, 223, 105169. https://doi.org/10.1016/j.compedu.2024.105169.

4. Personalized adaptive learning in higher education: A scoping review of key characteristics and impact on academic performance and engagement. (2024).

   Heliyon. https://doi.org/10.1016/j.heliyon.2024.e39630.

5. Artificial intelligence in personalized learning: A global systematic review of current advancements and shaping future opportunities. (2025). Social Sciences & Humanities Open, 12, 102114. https://doi.org/10.1016/j.ssaho.2025.102114.

6. Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education: Where are the educators? International Journal of Educational Technology in Higher Education, 16, 39. https://doi.org/10.1186/s41239-019-0171-0.

7. Shemshack, A., Kinshuk, & Spector, J. M. (2021). A comprehensive analysis of personalized learning components. Educational Technology Research and Development, 69, 179200.

8. Romero, C., & Ventura, S. (2020). Educational data mining and learning analytics: An updated survey. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 10(3), e1355.

9. Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43, 115135.

10. Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 3950.

11. Bandura, A. (1997). Self-Efficacy: The Exercise of Control. W. H. Freeman.

12. Deci, E. L., & Ryan, R. M. (2000). The “what” and “why” of goal pursuits: Human needs and the self-determination of behavior. Psychological Inquiry, 11(4), 227268.

13. Fredricks, J. A., Blumenfeld, P. C., & Paris, A. H. (2004). School engagement: Potential of the concept, state of the evidence. Review of Educational Research, 74(1), 59109.

14. Hillmayr, D., Ziernwald, L., Reinhold, F., Hofer, S. I., & Reiss, K. M. (2020). The potential of digital tools to enhance mathematics and science learning in secondary schools: A context-specific meta-analysis. Computers & Education, 153, 103897.

15. Luckin, R., Holmes, W., Griffiths, M., & Forcier, L. B. (2016). Intelligence Unleashed: An Argument for AI in Education. Pearson.

16. Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. Center for Curriculum Redesign.

17. Kline, R. B. (2016). Principles and Practice of Structural Equation Modeling (4th ed.). Guilford Press.

18. Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). Multivariate Data Analysis (8th ed.). Cengage.

19. UNESCO. (2023). Guidance for Generative AI in Education and Research. UNESCO.

20. Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & PRISMA Group. (2009). Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. BMJ, 339, b2535.