This paper proposed an online self-organizing radial basis function neural network (RBFNN) that dynamically adapts its structure and parameters based on input data distributions, employing Gaussian membership functions to determine neuron activation. The network autonomously adjusts its architecture by adding or removing neurons, ensuring relevance to the evolving data while maintaining computational efficiency. The Gaussian membership function for each neuron evaluates the similarity between input vectors and neuron centers, with the network output computed as a weighted sum of these activations. Furthermore, the self-organizing mechanism continuously monitors input patterns, introducing new neurons for novel data and pruning inactive ones, thereby optimizing the model’s representational capacity. The learning process iteratively updates neuron centers, widths, and weights via gradient descent to minimize prediction error, enabling the network to refine its performance incrementally. This approach addresses the limitations of static RBFNNs by providing adaptive learning capabilities, particularly suited for non-stationary environments where data distributions shift over time. The proposed method demonstrates significant potential in modeling complex input-output relationships, as evidenced by its ability to self-organize and generalize from streaming data. Moreover, the integration of dynamic Gaussian membership functions enhances the network’s flexibility, allowing it to capture intricate patterns without manual intervention. The results highlight the framework’s robustness and scalability, making it applicable to real-time learning tasks in domains such as time-series prediction, control systems, and adaptive filtering. By combining self-organization with online learning, our work advances the state-of-the-art in adaptive neural networks, offering a principled solution for scenarios requiring continuous model refinement.