A PCA/ICA based feature selection method and its application for corn fungi detection

Dimensionality reduction algorithms help reduce the classification time and sometimes the classification error. For time critical applications, in order to have reduction in the feature acquisition phase, feature selection is preferable to dimensionality reduction, which requires measurement of all inputs. Traditional feature selection methods, such as forward or backward selection, are costly to implement. We introduce a new feature selection method that decides on features to retain, based on how PCA (Principal Component Analysis) or ICA (Independent Component Analysis) values them. We compare the accuracy of our method to PCA and ICA using the same number of principal/independent components. We also do comparison to backward and forward selection with the same number of features. For our experiments, we use spectral measurement data taken from corn kernels infested and undamaged by fungi. Our algorithm selects features with almost as good classification accuracy as forward/backward selection and is a lot faster than those algorithms. It also results in better classification accuracy then using the same number of principal/independent components.