PaperlayerView exact revision on arXiv

Source-linked AI summary

A Hybrid AI Framework for Academic Advising: Integrating Ensemble-Based Grade Prediction and a Rule-Based Expert System

Hamid Saadatfar, Rohollah Hedayati-Nasab, AmirHossein Eshghi, Arash Hajihashemi

arXiv:2609.26243v1cs.AIcs.LG

TL;DR

Academic advising faces challenges as student populations grow, and the paper proposes a hybrid framework combining cluster-based grade prediction with rule-based recommendations. The ensemble reduces prediction error significantly, while the system’s stated limitations include a single-university dataset and exclusion of first-semester students.

  • Problem

    The framework addresses scalability, personalization, and transparency challenges in academic advising, where conventional approaches may focus on prediction or rule-based advising alone.

    Source: §5 · p. 24

  • Method

    The system clusters student records with GMM, trains cluster-specific grade-prediction models, and combines machine-learning predictions with a rule-based recommendation system.

    Source: §4.2 · p. 12, 5.Conclusion · pp. 26–27

  • Results

    The ensemble achieved the lowest MSE among the compared models, and its prediction-error reduction was statistically significant (paired-samples T-test, p < 0.001).

    Source: 4.Results · p. 15, §5.1 · p. 25

  • Takeaways & Limitations

    Rule-based explanations provide students with clear, actionable feedback that reduces cognitive ambiguity and increases confidence in the system’s guidance.

    Source: §5.2 · p. 25

    The dataset comes from a single university, so generalizability to other institutions requires further investigation.

    Source: §5.4 · p. 26

Abstract

from arXiv · show

The rapidly increasing student population has posed serious challenges to the traditional academic advising process. This study designs and implements a multi-purpose intelligent system to support students' academic progress, based on a two-part hybrid framework: (1) an advanced model for grade prediction and (2) a rule-based recommendation engine. Using a dataset containing 416,558 educational records from the University of Birjand, students were first divided into homogeneous clusters using the Gaussian Mixture Model (GMM). Subsequently, a Stacking Ensemble model combining Random Forest, Gradient Boosting, and MLP was trained specifically for each cluster. Evaluation results demonstrated that the Stacking model outperformed base models across all clusters, achieving a final aggregated RMSE of 2.35. The second component is an expert system that provides intelligent recommendations by synergizing educational regulations with the grades predicted by the first component. This system has been implemented as a practical tool on the University of Birjand portal, offering students real-time feedback such as semester GPA prediction, probation risk warnings, and course suggestions for GPA improvement.

2.Related work

Prior work explores AI-based advising through chatbots, course recommendations, and academic-performance prediction. The reviewed studies generally focus on one aspect, while this paper proposes integrating a clustering-based ensemble predictor with a rule-based expert system.

2.Related work

  • Course-recommendation studies combine curriculum and past performance, model course sequences, or recommend paths associated with successful graduates [12][13].The review also describes a framework that uses predicted performance and university regulations to recommend departments and courses.Source: 2.Related work · pp. 3–4
  • Grade-prediction studies use methods including Random Forest, knowledge tracing, deep learning, and Taylor-enhanced logistic regression, with reported performance varying across datasets and tasks [16][17][18][19][20].The studies report outcomes including accuracy, recall, F1-score, and interpretability; the review does not provide a common comparison across them.Source: 2.Related work · pp. 4–5
  • The reviewed studies mostly address prediction, recommendation, or chatbot support separately; this paper integrates a clustering-based ensemble predictor with a rule-based expert system.The review frames the integrated system as a response to the single-aspect focus of existing work.Source: 2.Related work · p. 5

3.Data and methodology

The methodology uses University of Birjand course records to develop a two-part advising system: cluster-specific grade prediction and rule-based course recommendations. It describes dataset analysis, ensemble learning, and hyperparameter selection.

3.1. Dataset

  • The University of Birjand student-portal dataset contains 416,558 records, each representing one course taken by a student, spanning the first semester of 2021–2022 through the first semester of 2024–2025.The study calculated Pearson correlations among feature pairs and between each feature and the target variable.Source: §3.1 · pp. 6–7

3.2. Framework

  • The proposed system combines a hybrid course-grade prediction framework with a rule-based engine for educational suggestions.Recommendations combine academic rules with predicted scores, suggesting courses for which students are eligible and likely to achieve a desired grade.Source: §3.2 · p. 9, Stage Two: Development of ensemble… · pp. 10–11
  • Gaussian Mixture Model clustering divides students into homogeneous groups, with the number of clusters selected through K-fold cross-validation.The method trains separate ensemble regression models for each cluster.Source: §3.2 · p. 9, Stage One: Data clustering for pattern… · p. 10

Stage Two: Development of ensemble prediction models

  • For each cluster, the ensemble combines Random Forest, Gradient Boosting, and MLP predictions using averaging to produce course-grade estimates.The stated aim is to use model diversity to reduce error and improve stability.Source: Stage Two: Development of ensemble… · p. 10
  • The rule-based system uses academic rules alone or combines them with predicted scores to suggest eligible courses with a high probability of the desired grade.The recommendations are intended for course registration in upcoming semesters.Source: Stage Two: Development of ensemble… · pp. 10–11

3.3. Hyperparameter Optimization

  • Grid Search with 5-fold cross-validation selects base-model hyperparameters by minimizing validation RMSE.Table 3 presents the search space and selected optimal values.Source: §3.3 · p. 11, Table 3 · p. 11

4.Results and Implementation

The results section presents the framework’s preprocessing, predictive performance, case study, and practical toolbox. The hybrid model clusters samples before training ensemble regressors for each group.

4.Results and Implementation

  • The results section also examines preprocessing, hybrid-model performance, a case study, and a practical grade-prediction and recommendation toolbox.Source: 4.Results and Implementation · p. 11

4.1. Preprocessing

  • Rows with zero or null values in key prior-GPA or similar-course columns were removed because those values indicated absence, not actual scores.Course, instructor, and major codes were encoded using mappings stored for consistent evaluation and deployment.Source: §4.1 · pp. 11–12
  • Numerical features were z-score standardized to mean zero and standard deviation one; training-set parameters were reused on validation and test data to prevent information leakage.For high-cardinality categories such as instructor ID, target means were smoothed with the global mean to reduce noise from categories with few samples.Source: §4.1 · p. 12

4.2. Creating a hybrid model

  • The framework uses GMM to identify similar statistical patterns, selecting ten clusters and applying clustering consistently to 80% training and 20% testing subsets with K-Fold cross-validation.The number of clusters was selected using BIC and AIC, which balance model fit and simplicity.Source: §4.2 · pp. 12–13
  • Each cluster receives a model averaging Random Forest, Gradient Boosting, and MLP predictions, producing ten final ensemble models.The MLP uses three hidden layers of 256, 128, and 64 neurons; the Random Forest and Gradient Boosting models each use 100 trees.Source: §4.2 · p. 13

4.Results

The results show that the ensemble outperformed the base models across clusters, followed by sample predictions, rule-based advising, portal deployment, and student usability feedback.

4.Results

  • Across clusters, the ensemble outperformed all base models on MSE, RMSE, and R²; Figure 6 compares R².The comparison covered results for each cluster and all reported metrics.Source: 4.Results · p. 15, Figure 6 · p. 16

4.1. Statistical Significance Analysis

  • The paired-samples t-test found a statistically significant reduction in ensemble prediction error on the unseen test set (p < 0.001).The analysis followed a Shapiro-Wilk normality check and used a 0.05 significance level.Source: §4.1 · p. 17

4.2. Sample Implementation

  • Three sampled students from clusters 4, 0, and 5 had predicted grade means of 10.62, 17.61, and 15.10, respectively.Their standard deviations were 2.18, 1.83, and 2.16, respectively, illustrating different predicted distributions and uncertainty across clusters.Source: §4.2 · p. 18

4.3. Rule-Based Recommendation System

  • The recommendation system matches student records against expert-defined rules to produce personalized warnings, recommendations, and motivational messages.It is designed to provide feedback during course registration and throughout students’ academic journeys.Source: §4.3 · p. 18

4.3.1. Detailed explanation of the implemented rules

  • Implemented rules check probation credit limits, unmet prerequisites, declining performance, critical predicted GPA, and courses that could improve GPA.Other rules flag slow academic progress and provide positive reinforcement when GPA improves.Source: §4.3.1 · p. 20

4.4. The developed tool implemented on the Birjand University integrated system

  • In the Birjand University portal, students can view predicted grades for selected courses, probable semester GPA, course suggestions, and rule-based alerts.Course suggestions use predicted grades to classify courses by difficulty and identify options with potential to improve GPA.Source: §4.4 · pp. 21–22

4.5. Usability Evaluation

  • Student feedback was positive: 82% said GPA predictions aided decisions, 88% affirmed probation-alert effectiveness, 76% found explanations clear and trustworthy, 88% rated navigation highly, and 90% wanted future use.Source: §4.5 · p. 24

5. Discussion

The discussion reports cluster-based predictive performance, explains how rule-based reasoning complements prediction, and describes deployment implications and study limitations.

5. Discussion

  • The study presents the framework as a practically deployable integration of ensemble learning and a transparent rule-based expert system for academic advising.The discussion frames the contribution around predictive performance, AI–expert-knowledge synergy, and practical implications.Source: §5 · pp. 24–25

5.1. Predictive Performance Analysis

  • The cluster-based ensemble outperformed base learners across clusters and metrics; on the unseen test set, it achieved RMSE 2.35 and R² 0.53.The paired-samples t-test yielded p < 0.001 under verified normality assumptions.Source: §5.1 · p. 25

5.2 Synergy of Artificial Intelligence and Expert Knowledge

  • GMM clustering creates homogeneous student subgroups for specialized predictors, while the rule-based engine enforces regulations, prerequisites, probation policies, and progression rules.The authors describe the rule layer as a transparent pedagogical guardrail complementary to performance prediction.Source: §5.1 · p. 25, §5.2 · p. 25

5.3. Practical Implications for Academic Advising

  • Portal deployment supports asynchronous advising, cluster-informed course personalization, and timely warnings about declining trends or probation risk before registration is finalized.The authors describe the system as available 24/7 and intended to leave faculty more time for complex psychological and career guidance.Source: §5.3 · p. 26

5.4. Limitations

  • The dataset comes from one university, so generalizability to other institutions remains to be evaluated; first-semester students were excluded for lacking prior records.The current system is therefore optimized for students with at least one semester of academic history.Source: §5.4 · p. 26

5.Conclusion

The study combines cluster-based grade prediction with a rule-based recommendation system deployed through a web platform. Future work will address the cold-start problem for freshmen using pre-admission data.

5.Conclusion

  • The rule-based engine supports policy compliance and gives end users transparent explanations.Source: 5.Conclusion · p. 27
  • The framework combines student-pattern clustering, specialized grade-prediction models, and a rule-based system for recommendations within a web-based platform.The prediction pipeline uses GMM clustering, Random Forest, Gradient Boosting, and MLP models, with averaging and stacking to aggregate outputs.Source: 5.Conclusion · pp. 26–27
  • Future work will address freshmen cold start by adding pre-admission data, including high school GPA and university entrance exam scores.Source: 5.Conclusion · p. 27

Statements and Declarations

Loading 2609.26243v1…