Machine learning in the boardroom: Gender diversity prediction using boosting and undersampling methods

Abstract

This paper addresses the crucial issue of boardroom diversity and proposes a novel approach utilizing machine learning to predict gender diversity on the boards of Chinese publicly-traded companies from 2008 to 2017. The study employs tree-based boosting with under-sampling as the machine learning technique. Various tree-based boosting techniques are utilized, and the evaluation is based on accuracy, precision, recall, F1 scores, and ROC scores. The findings reveal that extreme Gradient Boosting (XGBoost) with undersampling outperforms other models in terms of predictive performance. Moreover, the paper extracts interpretable principles in the form of if-else statements from the model to enhance its interpretability. This approach contributes to achieving corporate governance goals by promoting board gender diversity using machine learning techniques.