abstract
- Potentiometric electrochemical multi-sensors’ performance highly depends on the capability of selecting the best set of sensors. Indeed, signals are usually collinear resulting in over-fitted multivariate models with low predictive applicability. In this work, a comparative study was made to evaluate the predictive performance of classification models coupled with heuristic or metaheuristic variable selection algorithms. In this study, eleven single-cultivar extra virgin olive oils, from two crop years, were used. The results demonstrated that linear discriminant analysis with simulated annealing algorithm allowed selecting the best subset of sensors enabling 100% of correct cross-validation classifications, considering samples split by crop year.