Enhancing binary classification: A new stacking method via leveraging computational geometry

Binary classification is a fundamental task in machine learning and data science, with applications spanning numerous domains, including spam detection, medical diagnostics, image recognition, credit scoring. The goal is to predict a binary outcome--typically labeled as 0 or 1--based on a set of input features. Various machine learning algorithms, such as logistic regression (LR), k-nearest neighbors (kNN), support vector machines (SVM), and neural network (NN), are commonly employed for binary classification tasks. These algorithms can be mainly divided into two categories: those with interpretability, which are convenient for analysis and control (e.g., LR); and those without interpretability but with potentially good classification performance (e.g., NN). Ensemble learning, a powerful technique in predictive modeling, has gained widespread recognition for its ability to improve model performance by combining the strengths of multiple learning algorithms [1]. Among ensemble methods, stacking stands out by integrating the predictions of diverse base models (different learning algorithms) through a meta-model, resulting in enhanced prediction accuracy compared to only using the best base model [2]. Stacking has demonstrated significant applications in classification problems such as network intrusion detection [3, 4], cancer type classification [5], credit lending [6], and protein-protein binding affinity prediction [7].