Population Size Tradeoffs in De and Pso-based Methods for Pnn Training

We report on a comparative evaluation of three evolutionary methods for training the probabilistic neural network (PNN). The specific focus here is on an investigation of the acceptable tradeoffs, in terms of accuracy vs. computational and memory demands, depending on the population size. An empirical evaluation is carried out on the well-known Parkinson Speech Dataset with Multiple Types of Sound Recordings following a common experimental protocol. The numerical results identify the Unified PSO-based training as the most appropriate due to its superior classification accuracy and lower computational demands.