Assessment of the complementarity of data from multiple analytical techniques

Whilst the capabilities of analytical techniques are ever-increasing, individual methods can provide only a limited quantity of information about the composition of a complex mixture. Interrogation of samples by multiple techniques may permit for complementary information to be acquired, and suitable data fusion strategies are required in order to optimally exploit such complementary information. A novel mid-level data fusion strategy has been implemented which uses two-stage genetic programming for feature selection and canonical correlation analysis such that highly discriminatory variables can be related together in a multivariate fashion. The approach offers an intuitive way to visualise variable interaction and their contributions to experimental trends.