Nutt Tananimit, Thanatporn Grace, Yirga Abebe Belay, Rabian Wangkeeree, Compact gradient-based neural network for stochastic support vector regression with probabilistic constraints
Full Text: PDF
DOI: 10.23952/jnva.10.2026.6.04
Volume 10, Issue 6, 1 December 2026, Pages 1093-1116
Abstract. This paper introduces a neurodynamic system for solving stochastic support vector regression (SSVR) with probabilistic constraints. The proposed approach reformulates the SSVR problem into a deterministic quadratic programming problem. To address the nonsmoothness arising from the Karush–Kuhn–Tucker complementarity conditions, a smoothed Fischer–Burmeister function is employed, yielding a gradient-based neural network with enhanced numerical stability. The existence, uniqueness, and convergence of the network’s trajectories are rigorously analyzed, and the system is proven to be globally convergent to the optimal solution of the original SVR problem; Lyapunov stability is also established. Compared to existing methods, the proposed architecture achieves 25% reduction in network complexity leading to substantially faster training times. The efficiency and superiority of the model are validated through extensive numerical experiments on ten UCI benchmark datasets, where our method outperforms three existing SSVR baselines.
How to Cite this Article:
N. Tananimit, T. Grace, Y.A. Belay, R. Wangkeeree, Compact gradient-based neural network for stochastic support vector regression with probabilistic constraints, J. Nonlinear Var. Anal. 10 (2026), 1093-1116.