Centroid Neural Network With Directional Spawning of New Weights

This paper proposes Centroid Neural Network with Directional Spawning for efficient data clustering. The proposed algorithm locates each new neuron for cluster center by considering the location trajectory of previously added neurons during training process. Experiments on different data sets are performed in order to evaluate the performance of training error, training speed, and test error. The results show that the proposed algorithm outperforms conventional Centroid Neural Network and Fuzzy c-Means algorithm in terms of clustering errors and training speed. Keywords—Unsupervised learning algorithm, CNN, FCM, data clustering