A Comparative Analysis of Logistic Regression and Random Forest for Individual Fairness in Machine Learning

Abstract

In high-stakes domains such as finance, healthcare, and criminal justice, machine learning (ML) systems must balance predictive performance with fairness and transparency. This paper presents a comparative analysis of two widely used ML models, logistic regression and random forest, evaluated through the lens of individual fairness. Using the UCI Adult Income and COMPAS datasets, we assess performance in terms of accuracy, F1 score, individual consistency, and disparate treatment. Our findings indicate that while random forests offer marginally higher accuracy (by approximately 1%), logistic regression improves individual consistency by up to 4%, suggesting it is preferable in fairness-sensitive applications. This study emphasizes model selection’s role in achieving ethically responsible AI.