Prototype based low-dimensional kernels for interpretable classification

Interpretability in machine learning is crucial for building trust and understanding in model predictions, particularly in high-stakes domains, such as healthcare and finance. Linear models are considered interpretable due to their simplicity and transparency in representing relationships between input features and output predictions. However, their prediction performance may be limited in complex scenarios. To address this challenge, this research proposes using mapping functions based on prototypes to enhance prediction performance while preserving the interpretability of linear models. Prototypes are representative examples of the data that can be used to provide intuitive explanations for model predictions. The proposed mapping functions keeps original input features and, adds new features based on the distance to the prototypes, allowing linear models to capture complex relationships while maintaining interpretability. The proposed mapping functions are evaluated on various datasets, demonstrating improved prediction performance while maintaining interpretability. The use of prototypes provides intuitive explanations for model predictions, allowing practitioners to comprehend the final decisions made by the models. This work contributes to the field of machine learning by providing novel mapping functions to balancing prediction performance and interpretability in linear models