Diverse ensemble cost-sensitive logistic regression

In recent years, cost-sensitive methods have become increasingly crucial for decision-making in various real-world applications. These methods have been developed for the purpose of minimizing costs or risks for stakeholders. Moreover, the interpretability of cost-sensitive methods has gained considerable attention in critical domains such as finance and medical care. In this article, we propose a diverse ensemble of cost-sensitive logistic regression models to reduce costs for binary classification tasks, as well as a novel algorithm based on the partial conservative convex separable quadratic approximation to solve this non-convex optimization problem. The proposed method demonstrates substantial cost savings through extensive simulations and real-world applications, including fraud detection and gene expression analysis. Additionally, unlike other ensembling techniques, the resulting model of the proposed method is fully interpretable as a logistic regression model and achieves a high level of sparsity induced by the proposed algorithm. We believe this approach offers deeper insights into the relationship between predictors and response, enabling more informed decision-making in practical scenarios.