Wireless Sensor Networks (WSNs) are integral to various applications, demanding precise node localization. This paper presents a novel algorithm for WSN node localization, employing Gradient Descent Training (GDT). The algorithm takes as input the total number of WSN nodes, a subset of reference nodes with known coordinates, and training distance matrices estimated from initial random positions. Each node discovers its neighbour’s actual relative distances from received RF signal strength, contributing to the formation of a global distance matrix. It iteratively refines node positions to minimize the error between the original distance matrix and the training distance matrix. A key feature of the algorithm is the incorporation of re-initialization of the selected nodes with high error due to low signal to noise ratio to improve the error convergence. This convergence criterion overcomes local minima issues of GDT. Simulations demonstrate the effectiveness of the proposed algorithm in achieving accurate node localization within dynamic and noisy WSN environments. The approach’s potential to enhance real-world WSN applications is promising.