We present a framework to optimize the performance of a mobile ad hoc network over a wide range of operating conditions. It includes screening experiments to quantify the parameters and interactions among parameters influential to throughput. Profile-driven regression is applied to obtain a model of the non-linear behaviour of throughput. The intermediate models obtained in this modelling effort are used to adapt the parameters as the network conditions change, in order to maximize throughput. The improvements in throughput range from 10-26 times the use of the default parameter settings. The predictive accuracy of the model is monitored and used to update the model dynamically. The results indicate the framework may be useful for the optimization of dynamic systems of high dimension.