Online feature selection and classification

This paper presents an online feature selection and classification algorithm. The algorithm is implemented for impact acoustics signals to sort hazelnut kernels. The classifier, which is used to determine the most discriminative features, is updated when a new observation is processed. The algorithm starts with decomposing the signal both in time and frequency axes in binary tree format. A feature set is obtained from the extracted features by using each node of the trees in time-frequency (t-f) plane. The information gathered from new entrance is discarded after updating the model parameters and algorithm states. The binary trees are pruned both in time and frequency axes by using the discrimination power of the nodes. This gives the most discriminative sub-bands in the t-f axes. The relevant features are selected from the remaining nodes after pruning operation. A maximum likelihood classifier with the assumption of multivariate Gaussian distribution is obtained from the relevant model parameters, and used for online testing. The developed online learning algorithm gives better learning results compared to on-line AdaBoost algorithm for sorting of hazelnut kernels.