Evolutionary optimization studies are mainly focused on the type of methodologies and controlling parameters of the algorithms. However, the choice of fitness functions which gives a measure on how well a dataset correlates with reference values is often not justified in a rigorous way. The work in this paper addresses this issue by investigating the performance of a set of fitness functions and their suitability according to the input data used for training the optimization algorithms. A Genetic Algorithm was applied to a chemical kinetics problem with the aim of matching the measured species concentrations through optimization of a set of Arrhenius reaction rate parameters. The challenge with this type of problem lies in the wide range of species concentrations used as the training dataset, where they can be up to fifteen orders in magnitude. This revealed large differences in optimized values arising from the various fitness function formulations, thus highlighting the need for detailed knowledge of fitness functions for a specific optimization problem. Furthermore, the quality of training dataset through error propagation and a scaling parameter were assessed bringing further insight on differences fitness functions. Such information is crucial for the appropriate selection of fitness functions for the effective optimisation of reaction rate parameters. Crown Copyright (C) 2018 Published by Elsevier Ltd. All rights reserved.