Genetic algorithm based NARX model identification for evaluation of insulin sensitivity

The evaluation of insulin sensitivity plays an important role in the clinical investigation of glucose related diseases. Mathematical models based on non-invasive tests provide an estimate of insulin sensitivity by solving a nonlinear optimization problem. However traditional optimization methods suffer from convergence problem and the final estimate is heavily dependent on the initial parameterization. This paper proposes a model based on the hybrid approach of nonlinear autoregressive with exogenous input (NARX) modeling and genetic algorithm (GA) for deriving an index of insulin sensitivity. The model does not need an initial parameterization and the convergence is always guaranteed. The index derived from the proposed model is found to correlate well with the widely used minimal model based insulin sensitivity, with a significantly higher accuracy of fit.