• Prajakta Kolekar
• Dr. Ramdas Kolekar

Machine Learning (ML) systems are being used in critical decision-making systems, such as in educational admissions, in situations where this automated prediction may lead to irrevocable and long-term effects on both individual and societal lives. Although this is usually the case with such models, and the focus here is on predictive accuracy, high prediction does not necessarily imply fairness and disparities between groups at the group level may go undetected. This issue especially applies to the Indian education system, which can be described as having significant regional, social, and institutional diversity. Nevertheless, the majority of the available material on fairness is based on Western data or involves aggregated statistics and does not permit discrete-level fairness measurement in Indian admissions. This paper performs a fairness analysis of an individual-level synthetic model to appear like an Indian undergraduate admission on ML-based admission decisions on an individual-level synthetic college admission dataset. A baseline model of interpretability is used, namely the Logistic Regression, and a non-linear model with higher capacity as a baseline model is the Random Forest Classifier. This work is based on fairness evaluation and not mitigation of bias. Admission results are obtained by a fixed threshold of probability, and the notion of fairness using error-related criterion, namely False Negative Rate (FNR) and False Positive Rate (FPR), and more specifically False Negative Rate, since it is critically important to the admission decision. The assessment of fairness is done on such sensitive attributes as State, Gender, and Category. Both models have very high predictive accuracy in the experimental results, although small yet significant state-wide differences occur, whereas the gender and category-based differences are small. Notably, the more complex the model, the more unfairness is not found and, thus, the observed differences are mostly data-driven and not model-driven. It is the insight of these findings that fairness should be audited on a regular basis in the admission systems, and proves that being accurate can never guarantee fair decision-making in high-impact areas.

Algorithmic Fairness, Educational Admissions, Machine Learning Auditing, Indian Education System, Bias Evaluation

"Auditing State-wise Fairness in Machine Learning–Based Undergraduate Admissions: Evidence from an Indian Context ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 1, page no.b225-b233, January-2026, Available :http://www.jetir.org/papers/JETIR2601137.pdf

"Auditing State-wise Fairness in Machine Learning–Based Undergraduate Admissions: Evidence from an Indian Context ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 1, page no. ppb225-b233, January-2026, Available at : http://www.jetir.org/papers/JETIR2601137.pdf