Feature Selective Normalized Mutual Information Index for the validation of computational electromagnetics

An important concern in the validation of computational electromagnetics (CEM) is to incorporate quantitative indicators in order to perform conclusive comparisons between model predictions with respect to a reference expected response, such as data obtained from measurements, closed-form equations references or standard reference problems values. Besides, the aforementioned index must be qualitatively interpreted using established and objective criteria, mimicking the perception of individuals or teams of expert engineers and aiding them in the interpretation process. So far, this problem has been addressed by the standardized Feature Selective Validation (FSV) method which has become the most widespread technique in the validation of CEM. Nonetheless, the FSV has some drawbacks which have driven subsequent and non consolidated enhancements proposed to improve its performance under specific situations. This paper proposes a global and normalized mutual information based index that includes feature selective considerations as an alternative to evaluate and compare datasets involved in the validation of CEM. Finally, three numerical examples are developed using the presented methodology and the results are compared with the overall indicator for the amplitude and the feature measures obtained using the FSV. The results suggest that the proposed index is able to provide intuitive, consistent and compressive interpretation about the global results of the validation process. Nevertheless, research must be continued in order to refine the proposed method and testing its complete range of applicability and suitability.