An Improved Probabilistic Neural Network with GA Optimization
Huafen Yang, You Yang · 2012
Probabilistic Neural Network (PNN) was applied to prediction mainly. Over the traditional neural network, less time was cost by PNN in network architecture determining and training. But the smoothing parameter used in the estimation results was a user-defined constant. How to determine this parameter's value is a crucial problem in PNN. Combined with adaptive genetic algorithm (GA), a novel PNN was proposed. Crossover probability pcand mutation probability pmwere employed to optimize the smoothing parameter. These two probability factors were adaptive, they would vary with the fitness of population. A reasonable search breadth and depth was finished by the GA with pcand pm. The prediction accuracy of PNN was increased because of the smoothing parameters optimization. The simulation experiments demonstrated that the improved PNN model was effective. The prediction precision was increased more.