A genetic algorithm based on mutation and crossover with adaptive probabilities

We propose a probabilistic rule-based adaptive model (PRAM) where the mutation and the crossover rates are adapted dynamically throughout the running of genetic algorithms so that tedious parameter tuning can be avoided. Multi mutation and crossover rates are used for an epoch. A new set of rates is generated for the next epoch according to the fitness improvement. PRAM is compared with a commonly used benchmark adaptive strategy, self-adaptation, on a set of well-known numeric functions. Experimental results show that PRAM performs better than self-adaptation on both solution quality and efficiency