Fuzzy-logic system personalizes aesthetic education pathways for university students

A domain-specific fuzzy-logic system generated individualized aesthetic-education pathways and career-guidance recommendations for undergraduates at one comprehensive university in eastern China. In a study of 500 students, the model achieved 83.2% top-1 accuracy on a 100-student hold-out set, with a root-mean-square error of 0.176 and a mean absolute error of 0.141. The system’s main feature was not a new learning algorithm but an interpretable rule base connecting student characteristics to specific recommendations, together with a feedback mechanism designed to revise those recommendations as follow-up information became available.

The participants were full-time undergraduates drawn from arts and humanities, social sciences, engineering and natural sciences majors. Their grade distribution was 25% first-year, 28% second-year, 27% third-year and 20% fourth-year; 48% were male, 50% female and 2% selected other or preferred not to say. Researchers recorded participation in visual arts, music, theater and design workshops, satisfaction with aesthetic-education courses, career clarity, expected need for guidance, self-assessed ability, basic aesthetic literacy, personality openness and career-category preference. An a priori power analysis estimated that 386 participants would be needed for a medium effect size, and the final sample exceeded that threshold, although power was lower for some subgroup comparisons.

The survey instrument was developed through a literature review, review by five experts and a pilot involving 50 students who were excluded from the main sample. The expert panel rated item relevance, with items below the stated content-validity threshold revised or removed; the resulting instrument had an average scale content-validity index of 0.91. Pilot measures showed acceptable internal consistency for satisfaction, career clarity and guidance expectation, with Cronbach’s alpha values of 0.84, 0.79 and 0.76, respectively. Two-week test-retest correlations were 0.86 for participation frequency, 0.82 for satisfaction and 0.79 for career clarity. These checks establish consistency within this study population rather than the instrument’s validity across other universities or educational systems.

At the center of the model was a Mamdani-type inference engine designed to handle information that is graded rather than cleanly binary. Numerical inputs were normalized to the range from 0 to 1 and mapped to overlapping triangular membership functions labeled with terms such as low, medium and high. A satisfaction score of 7 out of 10, for example, could belong partly to moderately satisfied and partly to highly satisfied sets. Career clarity was treated in the same graduated way, with overlapping categories for unclear, moderate and clear states. Career-category preference was different: respondents’ interest-area text was mapped to one of six Holland occupational categories—artistic, social, investigative, conventional, enterprising or realistic—and manually checked. That categorical preference was not fuzzified and did not produce partial membership in several occupational categories; instead, it selected the rules applicable to a student.

At the center of the model was a Mamdani-type inference engine containing 55 IF-THEN rules. Each rule combined conditions such as participation frequency, satisfaction, aesthetic literacy, openness and career preference, and generated an output for recommendation strength or pathway type. The firing strength of a rule was determined by the weakest membership among its antecedents, while activated rules were aggregated with a maximum operation. Centroid defuzzification then converted the resulting fuzzy output into a continuous recommendation value. The system could display the three most strongly activated rules, allowing instructors to see which student attributes influenced a pathway and to provide a rationale alongside the output.

Five pathway types were described. Students with high participation, high satisfaction and clear interests were directed toward a specialized artistic track involving advanced studio practice and critique, typically six to eight hours per week. Those with strong engagement but somewhat less settled goals could receive an innovative fusion track combining arts with technology or business at four to six hours per week. Moderate engagement and career clarity produced an exploratory integrated track across several art forms, generally requiring three to four hours per week. High demand for guidance combined with moderate aesthetic literacy led to a practical skills-oriented track involving areas such as graphic design, public speaking and event planning, while low participation or unclear goals produced a foundational literacy track designed to build basic knowledge and skills at one to two hours per week.

Performance was generally stronger than that of the comparison models reported in the study. Average path-recommendation accuracy was 87%, compared with 71% for a decision tree and 69% for logistic regression. The fuzzy model’s mean squared error was 0.19, versus 0.34 and 0.29 for those comparators. Five-fold cross-validation produced an average score of 0.861, and a separate disturbance test found that output standard deviations remained within 0.021 when inputs varied by plus or minus 10%. In an expert comparison, the model’s recommendations agreed with manual judgments in 85.7% of cases. These figures describe performance within the sampled dataset; they do not establish that the approach will perform equally well in different institutional or cultural settings.

Removing one input illustrated why some information mattered for particular groups. Omitting aesthetic literacy increased RMSE from 0.176 to 0.197 and MAE from 0.141 to 0.158 among students with low prior exposure to the arts. Excluding personality openness increased RMSE by 8% for students with extreme openness scores, defined as below the 25th or above the 75th percentile. In the full model, high aesthetic literacy, high openness and an artistic career preference activated a rule recommending a creative-innovation pathway, whereas low literacy combined with a conventional career preference produced a foundational-literacy pathway. Such sensitivity analyses show that individual variables can influence particular recommendations, but they do not demonstrate that aesthetic education causes changes in career decisions or employment outcomes.

The researchers also designed a feedback loop for revising recommendations after students received them. Periodic measures of satisfaction and deviation between intended and actual development were used to adjust membership-function parameters and rule weights. In the illustrative case results, one student rated a pathway 9 out of 10 with a development-deviation score of 1, leading the system to maintain the original recommendation. Another had a deviation score of 4 and was moved from a foundational pathway toward exploration, including the addition of vocational-cognition courses. Across five student examples, pathway matching rose by 17% to 19%, intervention satisfaction increased by one to three points, and development deviation declined by two to four points after optimization. These examples are reported within-study observations and do not isolate the model’s effect from concurrent educational support or other influences.

Among the 100 students in the hold-out set who received initial recommendations, 72 reportedly followed their suggested path for at least eight weeks, and average path-fit satisfaction was 8.1 out of 10. The study also describes a modular implementation using Python, the skfuzzy library, NumPy, Matplotlib, Flask and ECharts, with an interface intended to return recommendations in less than half a second. The implementation is presented as a practical tool for educators, but the evidence comes from one institution, includes relatively few students in the longer follow-up analysis and does not include a randomized comparison of educational or career outcomes. The authors note that stable testing across comprehensive, vocational and art-focused institutions would require a multi-site sample of at least 1,200 participants, with subgroup-specific cross-validation.

The results position the system as domain-specific decision support rather than a new paradigm for artificial intelligence. Its appeal lies in connecting subjective student information to transparent recommendations without forcing every response into a sharp category, while allowing educators to inspect the rules behind each suggestion. Its central limitation is that a technically coherent and interpretable model is not the same as demonstrated educational benefit: the study evaluates recommendation accuracy, stability, satisfaction and selected follow-up outcomes, but it does not establish that participation in the recommended pathways causes better academic performance, better career decisions or better employment results. Larger multi-institution studies and more rigorous outcome evaluations are needed before the system’s educational value can be judged beyond the sampled setting.

Subject of Research: Evaluation of a fuzzy-logic system for individualized aesthetic-education pathways and career guidance among university students

Article Title: Path analysis of students’ aesthetic education and individualized construction of career guidance based on fuzzy logic

Article References: Zhang, G. (2026). Path analysis of students’ aesthetic education and individualized construction of career guidance based on fuzzy logic. Discover Artificial Intelligence, 6(1), Article 1038. https://doi.org/10.1007/s44163-026-01909-6

Image Credits: AI Generated

DOI: 10.1007/s44163-026-01909-6

Keywords: fuzzy logic, aesthetic education, career guidance, personalized learning, university students, Mamdani inference, decision support, path analysis, educational technology, Path, analysis, students