GAMLSS and neural networks in combat simulation metamodelling: A case study

• We compare additive regression models to neural networks in modelling a combat simulation.
• Additive models are better able to cope with skewed data.
• Out-of-sample performance is otherwise comparable.

The GAMLSS (Generalised Additive Models for Location, Scale and Shape) regression approach is compared to neural networks in the context of modelling the relationship between the inputs and outputs of the stochastic combat simulation model SIMBAT. The similarities and differences in these modelling approaches, and their advantages and disadvantages in this case, are discussed. Comparison of out-of-sample prediction suggests that some GAMLSS models are better able to cope with skewed data, but otherwise performance is broadly similar.