Analysis of algorithms and error evaluation criteria in chaotic neural network predictions

This article proposes several criteria for evaluating the performance of prediction algorithms, such as the root mean square error, bias, prediction accuracy, coefficient of determination, absolute error, and other normalized root mean square errors. The meanings of these criteria are then describe in detail. The two chaotic neural network prediction methods are then analyzed using the criteria. The results show that the global model algorithm has better performance than local model algorithm in chaotic neural network prediction and has a shorter training period, less source demand, and better generalization ability.