EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Data-Regime-Dependent Performance of Classical and Deep Learning Models for Psychometric Career-Interest Prediction

Minakshi Roy, Kalpana Sharma, Abhigyan Borah, Palash Ghosal

Artificial Intelligence and Applications · 15.11.2026

DOI: 10.47852/bonviewaia620211945 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Artificial Intelligence and Applications, online erschienen am 15.11.2026. Verfasst von Minakshi Roy und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate computational assessment of career interests can support educational and career counseling. However, psychometric models may differ in their ability to infer career-interest profiles and stated career preferences across different data regimes. The traditional methods of Realistic, Investigative, Artistic, Social, Enterprising, and Conventional (RIASEC) theory might not cover the nonlinear psychometric correlations. Classical machine learning and deep learning methods are considered for psychometric prediction of career interests and career preference classification in this paper, and a deep ensemble approach is suggested using multi-layer perceptrons (MLPs). Models are compared in terms of performance using two different datasets: Surabaya Career Dataset, which consists of more than 80,000 assessment records with O*NET-based RIASEC targets, and SMIT Engineering Career Aspirations Dataset, which consists of 1,863 personality-based profiles. In the Surabaya dataset, neural network models outperform the classical machine learning models by a margin; the scikit-learn MLP model has a high hit rate of 0.872. The deep ensemble model also has an almost similar hit rate of 0.850; this implies a small improvement over the single PyTorch MLP model hit rate of 0.849. In the case of the SMIT dataset, the performances are almost equal, with the deep ensemble hit rate being 0.404 while that of the XGBoost model being 0.403. These findings concern computational performance, not longitudinal career-outcome validity. Neural models have significant advantages in cases where psychometric data are large and rich in information, while classical models can compete with neural networks in the case of smaller data. Received: 31 August 2026 | Revised: 15 October 2026 | Accepted: 30 October 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Surabaya Career Dataset at OSF https://osf.io/27rbh/ and in SMIT Engineering Career Aspirations Dataset [Mendeley Data] at https://data.mendeley.com/datasets/66j432kt4w/1. Author Contribution Statement Minakshi Roy: Conceptualization, Methodology, Software, Validation, Investigation, Data curation, Writing – original draft, Visualization, Project administration. Kalpana Sharma: Formal analysis, Resources, Writing – review & editing, Supervision. Abhigyan Borah: Methodology, Software, Validation, Investigation, Data curation, Writing – original draft, Visualization. Palash Ghosal: Formal analysis, Resources, Writing – review & editing, Supervision.

Zitieren

APA

Roy, M., Sharma, K., Borah, A., & Ghosal, P. (2026). Data-Regime-Dependent Performance of Classical and Deep Learning Models for Psychometric Career-Interest Prediction. Artificial Intelligence and Applications. https://doi.org/10.47852/bonviewaia620211945

Vancouver

Roy M, Sharma K, Borah A, Ghosal P. Data-Regime-Dependent Performance of Classical and Deep Learning Models for Psychometric Career-Interest Prediction. Artificial Intelligence and Applications. 2026. doi:10.47852/bonviewaia620211945

BibTeX
@article{Roy2026,
  author = {Roy, Minakshi and Sharma, Kalpana and Borah, Abhigyan and Ghosal, Palash},
  title = {Data-Regime-Dependent Performance of Classical and Deep Learning Models for Psychometric Career-Interest Prediction},
  journal = {Artificial Intelligence and Applications},
  year = {2026},
  doi = {10.47852/bonviewaia620211945}
}
Kostenlos starten →