DNNSIM: Deep Monotonic Single-Index Model

This repository hosts supplementary materials and simulation code for the DNNSIM R package, available on CRAN. The package implements a monotonic single-index model using deep neural networks for robust modal regression, especially suited for skewed and heavy-tailed data. The preprint of this paper can be found here.

📄 Associated Paper

Title: A Monotonic Single-Index Model for Skewed and Heavy-Tailed Data: A Deep Neural Network Approach Applied to Periodontal Studies
Authors: Qingyang Liu, Shijie Wang, Ray Bai, Dipankar Bandyopadhyay
Abstract: This work proposes a deep monotonic single-index model with a flexible two-piece scale Student-(t) distribution, providing a robust alternative for analyzing skewed and heavy-tailed data. The proposed model ensures theoretical guarantees, outlier resistance, and interpretability via parametric index coefficients.

⚠️ Note: Due to size constraints, the real data (162MB) used in the paper’s application section is not included in this repository. Please contact the authors to request access.

🔗 CRAN Package

Install the latest release from CRAN:

install.packages("DNNSIM")