Neural network and least square estimator

Abstract

We propose an alternative and a methodology that joins the expressive power of neural networks with the robustness of least squares estimation to address regression problems involving a twice-censored setting. Within the context of modern statistical learning, this approach is particularly well suited for capturing complex and non-linear relationships. We establish mean squared error norm convergence of the proposed estimator and present a comparative simulation study demonstrating its superior performance over standard neural network methods.